1 /*
   2  * Copyright (c) 1999, 2021, 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/vmIntrinsics.hpp"
  29 #include "compiler/compileBroker.hpp"
  30 #include "compiler/compileLog.hpp"
  31 #include "gc/shared/barrierSet.hpp"
  32 #include "jfr/support/jfrIntrinsics.hpp"
  33 #include "memory/resourceArea.hpp"
  34 #include "oops/klass.inline.hpp"
  35 #include "oops/objArrayKlass.hpp"
  36 #include "opto/addnode.hpp"
  37 #include "opto/arraycopynode.hpp"
  38 #include "opto/c2compiler.hpp"
  39 #include "opto/castnode.hpp"
  40 #include "opto/cfgnode.hpp"
  41 #include "opto/convertnode.hpp"
  42 #include "opto/countbitsnode.hpp"
  43 #include "opto/idealKit.hpp"
  44 #include "opto/library_call.hpp"
  45 #include "opto/mathexactnode.hpp"
  46 #include "opto/mulnode.hpp"
  47 #include "opto/narrowptrnode.hpp"
  48 #include "opto/opaquenode.hpp"
  49 #include "opto/parse.hpp"
  50 #include "opto/runtime.hpp"
  51 #include "opto/rootnode.hpp"
  52 #include "opto/subnode.hpp"
  53 #include "prims/unsafe.hpp"
  54 #include "runtime/objectMonitor.hpp"
  55 #include "runtime/sharedRuntime.hpp"
  56 #include "runtime/stubRoutines.hpp"
  57 #include "utilities/macros.hpp"
  58 #include "utilities/powerOfTwo.hpp"
  59 
  60 #if INCLUDE_JFR
  61 #include "jfr/jfr.hpp"
  62 #endif
  63 
  64 //---------------------------make_vm_intrinsic----------------------------
  65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
  66   vmIntrinsicID id = m->intrinsic_id();
  67   assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
  68 
  69   if (!m->is_loaded()) {
  70     // Do not attempt to inline unloaded methods.
  71     return NULL;
  72   }
  73 
  74   C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
  75   bool is_available = false;
  76 
  77   {
  78     // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
  79     // the compiler must transition to '_thread_in_vm' state because both
  80     // methods access VM-internal data.
  81     VM_ENTRY_MARK;
  82     methodHandle mh(THREAD, m->get_Method());
  83     is_available = compiler != NULL && compiler->is_intrinsic_supported(mh, is_virtual) &&
  84                    !C->directive()->is_intrinsic_disabled(mh) &&
  85                    !vmIntrinsics::is_disabled_by_flags(mh);
  86 
  87   }
  88 
  89   if (is_available) {
  90     assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
  91     assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
  92     return new LibraryIntrinsic(m, is_virtual,
  93                                 vmIntrinsics::predicates_needed(id),
  94                                 vmIntrinsics::does_virtual_dispatch(id),
  95                                 id);
  96   } else {
  97     return NULL;
  98   }
  99 }
 100 
 101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
 102   LibraryCallKit kit(jvms, this);
 103   Compile* C = kit.C;
 104   int nodes = C->unique();
 105 #ifndef PRODUCT
 106   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 107     char buf[1000];
 108     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 109     tty->print_cr("Intrinsic %s", str);
 110   }
 111 #endif
 112   ciMethod* callee = kit.callee();
 113   const int bci    = kit.bci();
 114 #ifdef ASSERT
 115   Node* ctrl = kit.control();
 116 #endif
 117   // Try to inline the intrinsic.
 118   if (callee->check_intrinsic_candidate() &&
 119       kit.try_to_inline(_last_predicate)) {
 120     const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
 121                                           : "(intrinsic)";
 122     CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
 123     if (C->print_intrinsics() || C->print_inlining()) {
 124       C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
 125     }
 126     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 127     if (C->log()) {
 128       C->log()->elem("intrinsic id='%s'%s nodes='%d'",
 129                      vmIntrinsics::name_at(intrinsic_id()),
 130                      (is_virtual() ? " virtual='1'" : ""),
 131                      C->unique() - nodes);
 132     }
 133     // Push the result from the inlined method onto the stack.
 134     kit.push_result();
 135     C->print_inlining_update(this);
 136     return kit.transfer_exceptions_into_jvms();
 137   }
 138 
 139   // The intrinsic bailed out
 140   assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
 141   if (jvms->has_method()) {
 142     // Not a root compile.
 143     const char* msg;
 144     if (callee->intrinsic_candidate()) {
 145       msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
 146     } else {
 147       msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
 148                          : "failed to inline (intrinsic), method not annotated";
 149     }
 150     CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, msg);
 151     if (C->print_intrinsics() || C->print_inlining()) {
 152       C->print_inlining(callee, jvms->depth() - 1, bci, msg);
 153     }
 154   } else {
 155     // Root compile
 156     ResourceMark rm;
 157     stringStream msg_stream;
 158     msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
 159                      vmIntrinsics::name_at(intrinsic_id()),
 160                      is_virtual() ? " (virtual)" : "", bci);
 161     const char *msg = msg_stream.as_string();
 162     log_debug(jit, inlining)("%s", msg);
 163     if (C->print_intrinsics() || C->print_inlining()) {
 164       tty->print("%s", msg);
 165     }
 166   }
 167   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 168   C->print_inlining_update(this);
 169 
 170   return NULL;
 171 }
 172 
 173 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
 174   LibraryCallKit kit(jvms, this);
 175   Compile* C = kit.C;
 176   int nodes = C->unique();
 177   _last_predicate = predicate;
 178 #ifndef PRODUCT
 179   assert(is_predicated() && predicate < predicates_count(), "sanity");
 180   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
 181     char buf[1000];
 182     const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
 183     tty->print_cr("Predicate for intrinsic %s", str);
 184   }
 185 #endif
 186   ciMethod* callee = kit.callee();
 187   const int bci    = kit.bci();
 188 
 189   Node* slow_ctl = kit.try_to_predicate(predicate);
 190   if (!kit.failing()) {
 191     const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
 192                                           : "(intrinsic, predicate)";
 193     CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, inline_msg);
 194     if (C->print_intrinsics() || C->print_inlining()) {
 195       C->print_inlining(callee, jvms->depth() - 1, bci, inline_msg);
 196     }
 197     C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
 198     if (C->log()) {
 199       C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
 200                      vmIntrinsics::name_at(intrinsic_id()),
 201                      (is_virtual() ? " virtual='1'" : ""),
 202                      C->unique() - nodes);
 203     }
 204     return slow_ctl; // Could be NULL if the check folds.
 205   }
 206 
 207   // The intrinsic bailed out
 208   if (jvms->has_method()) {
 209     // Not a root compile.
 210     const char* msg = "failed to generate predicate for intrinsic";
 211     CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, msg);
 212     if (C->print_intrinsics() || C->print_inlining()) {
 213       C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
 214     }
 215   } else {
 216     // Root compile
 217     ResourceMark rm;
 218     stringStream msg_stream;
 219     msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
 220                      vmIntrinsics::name_at(intrinsic_id()),
 221                      is_virtual() ? " (virtual)" : "", bci);
 222     const char *msg = msg_stream.as_string();
 223     log_debug(jit, inlining)("%s", msg);
 224     if (C->print_intrinsics() || C->print_inlining()) {
 225       C->print_inlining_stream()->print("%s", msg);
 226     }
 227   }
 228   C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
 229   return NULL;
 230 }
 231 
 232 bool LibraryCallKit::try_to_inline(int predicate) {
 233   // Handle symbolic names for otherwise undistinguished boolean switches:
 234   const bool is_store       = true;
 235   const bool is_compress    = true;
 236   const bool is_static      = true;
 237   const bool is_volatile    = true;
 238 
 239   if (!jvms()->has_method()) {
 240     // Root JVMState has a null method.
 241     assert(map()->memory()->Opcode() == Op_Parm, "");
 242     // Insert the memory aliasing node
 243     set_all_memory(reset_memory());
 244   }
 245   assert(merged_memory(), "");
 246 
 247   switch (intrinsic_id()) {
 248   case vmIntrinsics::_hashCode:                 return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
 249   case vmIntrinsics::_identityHashCode:         return inline_native_hashcode(/*!virtual*/ false,         is_static);
 250   case vmIntrinsics::_getClass:                 return inline_native_getClass();
 251 
 252   case vmIntrinsics::_ceil:
 253   case vmIntrinsics::_floor:
 254   case vmIntrinsics::_rint:
 255   case vmIntrinsics::_dsin:
 256   case vmIntrinsics::_dcos:
 257   case vmIntrinsics::_dtan:
 258   case vmIntrinsics::_dabs:
 259   case vmIntrinsics::_fabs:
 260   case vmIntrinsics::_iabs:
 261   case vmIntrinsics::_labs:
 262   case vmIntrinsics::_datan2:
 263   case vmIntrinsics::_dsqrt:
 264   case vmIntrinsics::_dsqrt_strict:
 265   case vmIntrinsics::_dexp:
 266   case vmIntrinsics::_dlog:
 267   case vmIntrinsics::_dlog10:
 268   case vmIntrinsics::_dpow:
 269   case vmIntrinsics::_dcopySign:
 270   case vmIntrinsics::_fcopySign:
 271   case vmIntrinsics::_dsignum:
 272   case vmIntrinsics::_fsignum:                  return inline_math_native(intrinsic_id());
 273 
 274   case vmIntrinsics::_notify:
 275   case vmIntrinsics::_notifyAll:
 276     return inline_notify(intrinsic_id());
 277 
 278   case vmIntrinsics::_addExactI:                return inline_math_addExactI(false /* add */);
 279   case vmIntrinsics::_addExactL:                return inline_math_addExactL(false /* add */);
 280   case vmIntrinsics::_decrementExactI:          return inline_math_subtractExactI(true /* decrement */);
 281   case vmIntrinsics::_decrementExactL:          return inline_math_subtractExactL(true /* decrement */);
 282   case vmIntrinsics::_incrementExactI:          return inline_math_addExactI(true /* increment */);
 283   case vmIntrinsics::_incrementExactL:          return inline_math_addExactL(true /* increment */);
 284   case vmIntrinsics::_multiplyExactI:           return inline_math_multiplyExactI();
 285   case vmIntrinsics::_multiplyExactL:           return inline_math_multiplyExactL();
 286   case vmIntrinsics::_multiplyHigh:             return inline_math_multiplyHigh();
 287   case vmIntrinsics::_unsignedMultiplyHigh:     return inline_math_unsignedMultiplyHigh();
 288   case vmIntrinsics::_negateExactI:             return inline_math_negateExactI();
 289   case vmIntrinsics::_negateExactL:             return inline_math_negateExactL();
 290   case vmIntrinsics::_subtractExactI:           return inline_math_subtractExactI(false /* subtract */);
 291   case vmIntrinsics::_subtractExactL:           return inline_math_subtractExactL(false /* subtract */);
 292 
 293   case vmIntrinsics::_arraycopy:                return inline_arraycopy();
 294 
 295   case vmIntrinsics::_compareToL:               return inline_string_compareTo(StrIntrinsicNode::LL);
 296   case vmIntrinsics::_compareToU:               return inline_string_compareTo(StrIntrinsicNode::UU);
 297   case vmIntrinsics::_compareToLU:              return inline_string_compareTo(StrIntrinsicNode::LU);
 298   case vmIntrinsics::_compareToUL:              return inline_string_compareTo(StrIntrinsicNode::UL);
 299 
 300   case vmIntrinsics::_indexOfL:                 return inline_string_indexOf(StrIntrinsicNode::LL);
 301   case vmIntrinsics::_indexOfU:                 return inline_string_indexOf(StrIntrinsicNode::UU);
 302   case vmIntrinsics::_indexOfUL:                return inline_string_indexOf(StrIntrinsicNode::UL);
 303   case vmIntrinsics::_indexOfIL:                return inline_string_indexOfI(StrIntrinsicNode::LL);
 304   case vmIntrinsics::_indexOfIU:                return inline_string_indexOfI(StrIntrinsicNode::UU);
 305   case vmIntrinsics::_indexOfIUL:               return inline_string_indexOfI(StrIntrinsicNode::UL);
 306   case vmIntrinsics::_indexOfU_char:            return inline_string_indexOfChar(StrIntrinsicNode::U);
 307   case vmIntrinsics::_indexOfL_char:            return inline_string_indexOfChar(StrIntrinsicNode::L);
 308 
 309   case vmIntrinsics::_equalsL:                  return inline_string_equals(StrIntrinsicNode::LL);
 310   case vmIntrinsics::_equalsU:                  return inline_string_equals(StrIntrinsicNode::UU);
 311 
 312   case vmIntrinsics::_toBytesStringU:           return inline_string_toBytesU();
 313   case vmIntrinsics::_getCharsStringU:          return inline_string_getCharsU();
 314   case vmIntrinsics::_getCharStringU:           return inline_string_char_access(!is_store);
 315   case vmIntrinsics::_putCharStringU:           return inline_string_char_access( is_store);
 316 
 317   case vmIntrinsics::_compressStringC:
 318   case vmIntrinsics::_compressStringB:          return inline_string_copy( is_compress);
 319   case vmIntrinsics::_inflateStringC:
 320   case vmIntrinsics::_inflateStringB:           return inline_string_copy(!is_compress);
 321 
 322   case vmIntrinsics::_getReference:             return inline_unsafe_access(!is_store, T_OBJECT,   Relaxed, false);
 323   case vmIntrinsics::_getBoolean:               return inline_unsafe_access(!is_store, T_BOOLEAN,  Relaxed, false);
 324   case vmIntrinsics::_getByte:                  return inline_unsafe_access(!is_store, T_BYTE,     Relaxed, false);
 325   case vmIntrinsics::_getShort:                 return inline_unsafe_access(!is_store, T_SHORT,    Relaxed, false);
 326   case vmIntrinsics::_getChar:                  return inline_unsafe_access(!is_store, T_CHAR,     Relaxed, false);
 327   case vmIntrinsics::_getInt:                   return inline_unsafe_access(!is_store, T_INT,      Relaxed, false);
 328   case vmIntrinsics::_getLong:                  return inline_unsafe_access(!is_store, T_LONG,     Relaxed, false);
 329   case vmIntrinsics::_getFloat:                 return inline_unsafe_access(!is_store, T_FLOAT,    Relaxed, false);
 330   case vmIntrinsics::_getDouble:                return inline_unsafe_access(!is_store, T_DOUBLE,   Relaxed, false);
 331 
 332   case vmIntrinsics::_putReference:             return inline_unsafe_access( is_store, T_OBJECT,   Relaxed, false);
 333   case vmIntrinsics::_putBoolean:               return inline_unsafe_access( is_store, T_BOOLEAN,  Relaxed, false);
 334   case vmIntrinsics::_putByte:                  return inline_unsafe_access( is_store, T_BYTE,     Relaxed, false);
 335   case vmIntrinsics::_putShort:                 return inline_unsafe_access( is_store, T_SHORT,    Relaxed, false);
 336   case vmIntrinsics::_putChar:                  return inline_unsafe_access( is_store, T_CHAR,     Relaxed, false);
 337   case vmIntrinsics::_putInt:                   return inline_unsafe_access( is_store, T_INT,      Relaxed, false);
 338   case vmIntrinsics::_putLong:                  return inline_unsafe_access( is_store, T_LONG,     Relaxed, false);
 339   case vmIntrinsics::_putFloat:                 return inline_unsafe_access( is_store, T_FLOAT,    Relaxed, false);
 340   case vmIntrinsics::_putDouble:                return inline_unsafe_access( is_store, T_DOUBLE,   Relaxed, false);
 341 
 342   case vmIntrinsics::_getReferenceVolatile:     return inline_unsafe_access(!is_store, T_OBJECT,   Volatile, false);
 343   case vmIntrinsics::_getBooleanVolatile:       return inline_unsafe_access(!is_store, T_BOOLEAN,  Volatile, false);
 344   case vmIntrinsics::_getByteVolatile:          return inline_unsafe_access(!is_store, T_BYTE,     Volatile, false);
 345   case vmIntrinsics::_getShortVolatile:         return inline_unsafe_access(!is_store, T_SHORT,    Volatile, false);
 346   case vmIntrinsics::_getCharVolatile:          return inline_unsafe_access(!is_store, T_CHAR,     Volatile, false);
 347   case vmIntrinsics::_getIntVolatile:           return inline_unsafe_access(!is_store, T_INT,      Volatile, false);
 348   case vmIntrinsics::_getLongVolatile:          return inline_unsafe_access(!is_store, T_LONG,     Volatile, false);
 349   case vmIntrinsics::_getFloatVolatile:         return inline_unsafe_access(!is_store, T_FLOAT,    Volatile, false);
 350   case vmIntrinsics::_getDoubleVolatile:        return inline_unsafe_access(!is_store, T_DOUBLE,   Volatile, false);
 351 
 352   case vmIntrinsics::_putReferenceVolatile:     return inline_unsafe_access( is_store, T_OBJECT,   Volatile, false);
 353   case vmIntrinsics::_putBooleanVolatile:       return inline_unsafe_access( is_store, T_BOOLEAN,  Volatile, false);
 354   case vmIntrinsics::_putByteVolatile:          return inline_unsafe_access( is_store, T_BYTE,     Volatile, false);
 355   case vmIntrinsics::_putShortVolatile:         return inline_unsafe_access( is_store, T_SHORT,    Volatile, false);
 356   case vmIntrinsics::_putCharVolatile:          return inline_unsafe_access( is_store, T_CHAR,     Volatile, false);
 357   case vmIntrinsics::_putIntVolatile:           return inline_unsafe_access( is_store, T_INT,      Volatile, false);
 358   case vmIntrinsics::_putLongVolatile:          return inline_unsafe_access( is_store, T_LONG,     Volatile, false);
 359   case vmIntrinsics::_putFloatVolatile:         return inline_unsafe_access( is_store, T_FLOAT,    Volatile, false);
 360   case vmIntrinsics::_putDoubleVolatile:        return inline_unsafe_access( is_store, T_DOUBLE,   Volatile, false);
 361 
 362   case vmIntrinsics::_getShortUnaligned:        return inline_unsafe_access(!is_store, T_SHORT,    Relaxed, true);
 363   case vmIntrinsics::_getCharUnaligned:         return inline_unsafe_access(!is_store, T_CHAR,     Relaxed, true);
 364   case vmIntrinsics::_getIntUnaligned:          return inline_unsafe_access(!is_store, T_INT,      Relaxed, true);
 365   case vmIntrinsics::_getLongUnaligned:         return inline_unsafe_access(!is_store, T_LONG,     Relaxed, true);
 366 
 367   case vmIntrinsics::_putShortUnaligned:        return inline_unsafe_access( is_store, T_SHORT,    Relaxed, true);
 368   case vmIntrinsics::_putCharUnaligned:         return inline_unsafe_access( is_store, T_CHAR,     Relaxed, true);
 369   case vmIntrinsics::_putIntUnaligned:          return inline_unsafe_access( is_store, T_INT,      Relaxed, true);
 370   case vmIntrinsics::_putLongUnaligned:         return inline_unsafe_access( is_store, T_LONG,     Relaxed, true);
 371 
 372   case vmIntrinsics::_getReferenceAcquire:      return inline_unsafe_access(!is_store, T_OBJECT,   Acquire, false);
 373   case vmIntrinsics::_getBooleanAcquire:        return inline_unsafe_access(!is_store, T_BOOLEAN,  Acquire, false);
 374   case vmIntrinsics::_getByteAcquire:           return inline_unsafe_access(!is_store, T_BYTE,     Acquire, false);
 375   case vmIntrinsics::_getShortAcquire:          return inline_unsafe_access(!is_store, T_SHORT,    Acquire, false);
 376   case vmIntrinsics::_getCharAcquire:           return inline_unsafe_access(!is_store, T_CHAR,     Acquire, false);
 377   case vmIntrinsics::_getIntAcquire:            return inline_unsafe_access(!is_store, T_INT,      Acquire, false);
 378   case vmIntrinsics::_getLongAcquire:           return inline_unsafe_access(!is_store, T_LONG,     Acquire, false);
 379   case vmIntrinsics::_getFloatAcquire:          return inline_unsafe_access(!is_store, T_FLOAT,    Acquire, false);
 380   case vmIntrinsics::_getDoubleAcquire:         return inline_unsafe_access(!is_store, T_DOUBLE,   Acquire, false);
 381 
 382   case vmIntrinsics::_putReferenceRelease:      return inline_unsafe_access( is_store, T_OBJECT,   Release, false);
 383   case vmIntrinsics::_putBooleanRelease:        return inline_unsafe_access( is_store, T_BOOLEAN,  Release, false);
 384   case vmIntrinsics::_putByteRelease:           return inline_unsafe_access( is_store, T_BYTE,     Release, false);
 385   case vmIntrinsics::_putShortRelease:          return inline_unsafe_access( is_store, T_SHORT,    Release, false);
 386   case vmIntrinsics::_putCharRelease:           return inline_unsafe_access( is_store, T_CHAR,     Release, false);
 387   case vmIntrinsics::_putIntRelease:            return inline_unsafe_access( is_store, T_INT,      Release, false);
 388   case vmIntrinsics::_putLongRelease:           return inline_unsafe_access( is_store, T_LONG,     Release, false);
 389   case vmIntrinsics::_putFloatRelease:          return inline_unsafe_access( is_store, T_FLOAT,    Release, false);
 390   case vmIntrinsics::_putDoubleRelease:         return inline_unsafe_access( is_store, T_DOUBLE,   Release, false);
 391 
 392   case vmIntrinsics::_getReferenceOpaque:       return inline_unsafe_access(!is_store, T_OBJECT,   Opaque, false);
 393   case vmIntrinsics::_getBooleanOpaque:         return inline_unsafe_access(!is_store, T_BOOLEAN,  Opaque, false);
 394   case vmIntrinsics::_getByteOpaque:            return inline_unsafe_access(!is_store, T_BYTE,     Opaque, false);
 395   case vmIntrinsics::_getShortOpaque:           return inline_unsafe_access(!is_store, T_SHORT,    Opaque, false);
 396   case vmIntrinsics::_getCharOpaque:            return inline_unsafe_access(!is_store, T_CHAR,     Opaque, false);
 397   case vmIntrinsics::_getIntOpaque:             return inline_unsafe_access(!is_store, T_INT,      Opaque, false);
 398   case vmIntrinsics::_getLongOpaque:            return inline_unsafe_access(!is_store, T_LONG,     Opaque, false);
 399   case vmIntrinsics::_getFloatOpaque:           return inline_unsafe_access(!is_store, T_FLOAT,    Opaque, false);
 400   case vmIntrinsics::_getDoubleOpaque:          return inline_unsafe_access(!is_store, T_DOUBLE,   Opaque, false);
 401 
 402   case vmIntrinsics::_putReferenceOpaque:       return inline_unsafe_access( is_store, T_OBJECT,   Opaque, false);
 403   case vmIntrinsics::_putBooleanOpaque:         return inline_unsafe_access( is_store, T_BOOLEAN,  Opaque, false);
 404   case vmIntrinsics::_putByteOpaque:            return inline_unsafe_access( is_store, T_BYTE,     Opaque, false);
 405   case vmIntrinsics::_putShortOpaque:           return inline_unsafe_access( is_store, T_SHORT,    Opaque, false);
 406   case vmIntrinsics::_putCharOpaque:            return inline_unsafe_access( is_store, T_CHAR,     Opaque, false);
 407   case vmIntrinsics::_putIntOpaque:             return inline_unsafe_access( is_store, T_INT,      Opaque, false);
 408   case vmIntrinsics::_putLongOpaque:            return inline_unsafe_access( is_store, T_LONG,     Opaque, false);
 409   case vmIntrinsics::_putFloatOpaque:           return inline_unsafe_access( is_store, T_FLOAT,    Opaque, false);
 410   case vmIntrinsics::_putDoubleOpaque:          return inline_unsafe_access( is_store, T_DOUBLE,   Opaque, false);
 411 
 412   case vmIntrinsics::_compareAndSetReference:   return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap,      Volatile);
 413   case vmIntrinsics::_compareAndSetByte:        return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap,      Volatile);
 414   case vmIntrinsics::_compareAndSetShort:       return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap,      Volatile);
 415   case vmIntrinsics::_compareAndSetInt:         return inline_unsafe_load_store(T_INT,    LS_cmp_swap,      Volatile);
 416   case vmIntrinsics::_compareAndSetLong:        return inline_unsafe_load_store(T_LONG,   LS_cmp_swap,      Volatile);
 417 
 418   case vmIntrinsics::_weakCompareAndSetReferencePlain:     return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
 419   case vmIntrinsics::_weakCompareAndSetReferenceAcquire:   return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
 420   case vmIntrinsics::_weakCompareAndSetReferenceRelease:   return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
 421   case vmIntrinsics::_weakCompareAndSetReference:          return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
 422   case vmIntrinsics::_weakCompareAndSetBytePlain:          return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Relaxed);
 423   case vmIntrinsics::_weakCompareAndSetByteAcquire:        return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Acquire);
 424   case vmIntrinsics::_weakCompareAndSetByteRelease:        return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Release);
 425   case vmIntrinsics::_weakCompareAndSetByte:               return inline_unsafe_load_store(T_BYTE,   LS_cmp_swap_weak, Volatile);
 426   case vmIntrinsics::_weakCompareAndSetShortPlain:         return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Relaxed);
 427   case vmIntrinsics::_weakCompareAndSetShortAcquire:       return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Acquire);
 428   case vmIntrinsics::_weakCompareAndSetShortRelease:       return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Release);
 429   case vmIntrinsics::_weakCompareAndSetShort:              return inline_unsafe_load_store(T_SHORT,  LS_cmp_swap_weak, Volatile);
 430   case vmIntrinsics::_weakCompareAndSetIntPlain:           return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Relaxed);
 431   case vmIntrinsics::_weakCompareAndSetIntAcquire:         return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Acquire);
 432   case vmIntrinsics::_weakCompareAndSetIntRelease:         return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Release);
 433   case vmIntrinsics::_weakCompareAndSetInt:                return inline_unsafe_load_store(T_INT,    LS_cmp_swap_weak, Volatile);
 434   case vmIntrinsics::_weakCompareAndSetLongPlain:          return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Relaxed);
 435   case vmIntrinsics::_weakCompareAndSetLongAcquire:        return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Acquire);
 436   case vmIntrinsics::_weakCompareAndSetLongRelease:        return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Release);
 437   case vmIntrinsics::_weakCompareAndSetLong:               return inline_unsafe_load_store(T_LONG,   LS_cmp_swap_weak, Volatile);
 438 
 439   case vmIntrinsics::_compareAndExchangeReference:         return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Volatile);
 440   case vmIntrinsics::_compareAndExchangeReferenceAcquire:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Acquire);
 441   case vmIntrinsics::_compareAndExchangeReferenceRelease:  return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange,  Release);
 442   case vmIntrinsics::_compareAndExchangeByte:              return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Volatile);
 443   case vmIntrinsics::_compareAndExchangeByteAcquire:       return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Acquire);
 444   case vmIntrinsics::_compareAndExchangeByteRelease:       return inline_unsafe_load_store(T_BYTE,   LS_cmp_exchange,  Release);
 445   case vmIntrinsics::_compareAndExchangeShort:             return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Volatile);
 446   case vmIntrinsics::_compareAndExchangeShortAcquire:      return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Acquire);
 447   case vmIntrinsics::_compareAndExchangeShortRelease:      return inline_unsafe_load_store(T_SHORT,  LS_cmp_exchange,  Release);
 448   case vmIntrinsics::_compareAndExchangeInt:               return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Volatile);
 449   case vmIntrinsics::_compareAndExchangeIntAcquire:        return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Acquire);
 450   case vmIntrinsics::_compareAndExchangeIntRelease:        return inline_unsafe_load_store(T_INT,    LS_cmp_exchange,  Release);
 451   case vmIntrinsics::_compareAndExchangeLong:              return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Volatile);
 452   case vmIntrinsics::_compareAndExchangeLongAcquire:       return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Acquire);
 453   case vmIntrinsics::_compareAndExchangeLongRelease:       return inline_unsafe_load_store(T_LONG,   LS_cmp_exchange,  Release);
 454 
 455   case vmIntrinsics::_getAndAddByte:                    return inline_unsafe_load_store(T_BYTE,   LS_get_add,       Volatile);
 456   case vmIntrinsics::_getAndAddShort:                   return inline_unsafe_load_store(T_SHORT,  LS_get_add,       Volatile);
 457   case vmIntrinsics::_getAndAddInt:                     return inline_unsafe_load_store(T_INT,    LS_get_add,       Volatile);
 458   case vmIntrinsics::_getAndAddLong:                    return inline_unsafe_load_store(T_LONG,   LS_get_add,       Volatile);
 459 
 460   case vmIntrinsics::_getAndSetByte:                    return inline_unsafe_load_store(T_BYTE,   LS_get_set,       Volatile);
 461   case vmIntrinsics::_getAndSetShort:                   return inline_unsafe_load_store(T_SHORT,  LS_get_set,       Volatile);
 462   case vmIntrinsics::_getAndSetInt:                     return inline_unsafe_load_store(T_INT,    LS_get_set,       Volatile);
 463   case vmIntrinsics::_getAndSetLong:                    return inline_unsafe_load_store(T_LONG,   LS_get_set,       Volatile);
 464   case vmIntrinsics::_getAndSetReference:               return inline_unsafe_load_store(T_OBJECT, LS_get_set,       Volatile);
 465 
 466   case vmIntrinsics::_loadFence:
 467   case vmIntrinsics::_storeFence:
 468   case vmIntrinsics::_storeStoreFence:
 469   case vmIntrinsics::_fullFence:                return inline_unsafe_fence(intrinsic_id());
 470 
 471   case vmIntrinsics::_onSpinWait:               return inline_onspinwait();
 472 
 473   case vmIntrinsics::_currentThread:            return inline_native_currentThread();
 474 
 475 #ifdef JFR_HAVE_INTRINSICS
 476   case vmIntrinsics::_counterTime:              return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JFR_TIME_FUNCTION), "counterTime");
 477   case vmIntrinsics::_getClassId:               return inline_native_classID();
 478   case vmIntrinsics::_getEventWriter:           return inline_native_getEventWriter();
 479 #endif
 480   case vmIntrinsics::_currentTimeMillis:        return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
 481   case vmIntrinsics::_nanoTime:                 return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
 482   case vmIntrinsics::_writeback0:               return inline_unsafe_writeback0();
 483   case vmIntrinsics::_writebackPreSync0:        return inline_unsafe_writebackSync0(true);
 484   case vmIntrinsics::_writebackPostSync0:       return inline_unsafe_writebackSync0(false);
 485   case vmIntrinsics::_allocateInstance:         return inline_unsafe_allocate();
 486   case vmIntrinsics::_copyMemory:               return inline_unsafe_copyMemory();
 487   case vmIntrinsics::_getLength:                return inline_native_getLength();
 488   case vmIntrinsics::_copyOf:                   return inline_array_copyOf(false);
 489   case vmIntrinsics::_copyOfRange:              return inline_array_copyOf(true);
 490   case vmIntrinsics::_equalsB:                  return inline_array_equals(StrIntrinsicNode::LL);
 491   case vmIntrinsics::_equalsC:                  return inline_array_equals(StrIntrinsicNode::UU);
 492   case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
 493   case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
 494   case vmIntrinsics::_clone:                    return inline_native_clone(intrinsic()->is_virtual());
 495 
 496   case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
 497   case vmIntrinsics::_newArray:                   return inline_unsafe_newArray(false);
 498 
 499   case vmIntrinsics::_isAssignableFrom:         return inline_native_subtype_check();
 500 
 501   case vmIntrinsics::_isInstance:
 502   case vmIntrinsics::_getModifiers:
 503   case vmIntrinsics::_isInterface:
 504   case vmIntrinsics::_isArray:
 505   case vmIntrinsics::_isPrimitive:
 506   case vmIntrinsics::_isHidden:
 507   case vmIntrinsics::_getSuperclass:
 508   case vmIntrinsics::_getClassAccessFlags:      return inline_native_Class_query(intrinsic_id());
 509 
 510   case vmIntrinsics::_floatToRawIntBits:
 511   case vmIntrinsics::_floatToIntBits:
 512   case vmIntrinsics::_intBitsToFloat:
 513   case vmIntrinsics::_doubleToRawLongBits:
 514   case vmIntrinsics::_doubleToLongBits:
 515   case vmIntrinsics::_longBitsToDouble:         return inline_fp_conversions(intrinsic_id());
 516 
 517   case vmIntrinsics::_numberOfLeadingZeros_i:
 518   case vmIntrinsics::_numberOfLeadingZeros_l:
 519   case vmIntrinsics::_numberOfTrailingZeros_i:
 520   case vmIntrinsics::_numberOfTrailingZeros_l:
 521   case vmIntrinsics::_bitCount_i:
 522   case vmIntrinsics::_bitCount_l:
 523   case vmIntrinsics::_reverseBytes_i:
 524   case vmIntrinsics::_reverseBytes_l:
 525   case vmIntrinsics::_reverseBytes_s:
 526   case vmIntrinsics::_reverseBytes_c:           return inline_number_methods(intrinsic_id());
 527 
 528   case vmIntrinsics::_getCallerClass:           return inline_native_Reflection_getCallerClass();
 529 
 530   case vmIntrinsics::_Reference_get:            return inline_reference_get();
 531   case vmIntrinsics::_Reference_refersTo0:      return inline_reference_refersTo0(false);
 532   case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
 533 
 534   case vmIntrinsics::_Class_cast:               return inline_Class_cast();
 535 
 536   case vmIntrinsics::_aescrypt_encryptBlock:
 537   case vmIntrinsics::_aescrypt_decryptBlock:    return inline_aescrypt_Block(intrinsic_id());
 538 
 539   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 540   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 541     return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
 542 
 543   case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
 544   case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
 545     return inline_electronicCodeBook_AESCrypt(intrinsic_id());
 546 
 547   case vmIntrinsics::_counterMode_AESCrypt:
 548     return inline_counterMode_AESCrypt(intrinsic_id());
 549 
 550   case vmIntrinsics::_galoisCounterMode_AESCrypt:
 551     return inline_galoisCounterMode_AESCrypt();
 552 
 553   case vmIntrinsics::_md5_implCompress:
 554   case vmIntrinsics::_sha_implCompress:
 555   case vmIntrinsics::_sha2_implCompress:
 556   case vmIntrinsics::_sha5_implCompress:
 557   case vmIntrinsics::_sha3_implCompress:
 558     return inline_digestBase_implCompress(intrinsic_id());
 559 
 560   case vmIntrinsics::_digestBase_implCompressMB:
 561     return inline_digestBase_implCompressMB(predicate);
 562 
 563   case vmIntrinsics::_multiplyToLen:
 564     return inline_multiplyToLen();
 565 
 566   case vmIntrinsics::_squareToLen:
 567     return inline_squareToLen();
 568 
 569   case vmIntrinsics::_mulAdd:
 570     return inline_mulAdd();
 571 
 572   case vmIntrinsics::_montgomeryMultiply:
 573     return inline_montgomeryMultiply();
 574   case vmIntrinsics::_montgomerySquare:
 575     return inline_montgomerySquare();
 576 
 577   case vmIntrinsics::_bigIntegerRightShiftWorker:
 578     return inline_bigIntegerShift(true);
 579   case vmIntrinsics::_bigIntegerLeftShiftWorker:
 580     return inline_bigIntegerShift(false);
 581 
 582   case vmIntrinsics::_vectorizedMismatch:
 583     return inline_vectorizedMismatch();
 584 
 585   case vmIntrinsics::_ghash_processBlocks:
 586     return inline_ghash_processBlocks();
 587   case vmIntrinsics::_base64_encodeBlock:
 588     return inline_base64_encodeBlock();
 589   case vmIntrinsics::_base64_decodeBlock:
 590     return inline_base64_decodeBlock();
 591 
 592   case vmIntrinsics::_encodeISOArray:
 593   case vmIntrinsics::_encodeByteISOArray:
 594     return inline_encodeISOArray(false);
 595   case vmIntrinsics::_encodeAsciiArray:
 596     return inline_encodeISOArray(true);
 597 
 598   case vmIntrinsics::_updateCRC32:
 599     return inline_updateCRC32();
 600   case vmIntrinsics::_updateBytesCRC32:
 601     return inline_updateBytesCRC32();
 602   case vmIntrinsics::_updateByteBufferCRC32:
 603     return inline_updateByteBufferCRC32();
 604 
 605   case vmIntrinsics::_updateBytesCRC32C:
 606     return inline_updateBytesCRC32C();
 607   case vmIntrinsics::_updateDirectByteBufferCRC32C:
 608     return inline_updateDirectByteBufferCRC32C();
 609 
 610   case vmIntrinsics::_updateBytesAdler32:
 611     return inline_updateBytesAdler32();
 612   case vmIntrinsics::_updateByteBufferAdler32:
 613     return inline_updateByteBufferAdler32();
 614 
 615   case vmIntrinsics::_profileBoolean:
 616     return inline_profileBoolean();
 617   case vmIntrinsics::_isCompileConstant:
 618     return inline_isCompileConstant();
 619 
 620   case vmIntrinsics::_hasNegatives:
 621     return inline_hasNegatives();
 622 
 623   case vmIntrinsics::_fmaD:
 624   case vmIntrinsics::_fmaF:
 625     return inline_fma(intrinsic_id());
 626 
 627   case vmIntrinsics::_isDigit:
 628   case vmIntrinsics::_isLowerCase:
 629   case vmIntrinsics::_isUpperCase:
 630   case vmIntrinsics::_isWhitespace:
 631     return inline_character_compare(intrinsic_id());
 632 
 633   case vmIntrinsics::_min:
 634   case vmIntrinsics::_max:
 635   case vmIntrinsics::_min_strict:
 636   case vmIntrinsics::_max_strict:
 637     return inline_min_max(intrinsic_id());
 638 
 639   case vmIntrinsics::_maxF:
 640   case vmIntrinsics::_minF:
 641   case vmIntrinsics::_maxD:
 642   case vmIntrinsics::_minD:
 643   case vmIntrinsics::_maxF_strict:
 644   case vmIntrinsics::_minF_strict:
 645   case vmIntrinsics::_maxD_strict:
 646   case vmIntrinsics::_minD_strict:
 647       return inline_fp_min_max(intrinsic_id());
 648 
 649   case vmIntrinsics::_VectorUnaryOp:
 650     return inline_vector_nary_operation(1);
 651   case vmIntrinsics::_VectorBinaryOp:
 652     return inline_vector_nary_operation(2);
 653   case vmIntrinsics::_VectorTernaryOp:
 654     return inline_vector_nary_operation(3);
 655   case vmIntrinsics::_VectorFromBitsCoerced:
 656     return inline_vector_frombits_coerced();
 657   case vmIntrinsics::_VectorShuffleIota:
 658     return inline_vector_shuffle_iota();
 659   case vmIntrinsics::_VectorMaskOp:
 660     return inline_vector_mask_operation();
 661   case vmIntrinsics::_VectorShuffleToVector:
 662     return inline_vector_shuffle_to_vector();
 663   case vmIntrinsics::_VectorLoadOp:
 664     return inline_vector_mem_operation(/*is_store=*/false);
 665   case vmIntrinsics::_VectorLoadMaskedOp:
 666     return inline_vector_mem_masked_operation(/*is_store*/false);
 667   case vmIntrinsics::_VectorStoreOp:
 668     return inline_vector_mem_operation(/*is_store=*/true);
 669   case vmIntrinsics::_VectorStoreMaskedOp:
 670     return inline_vector_mem_masked_operation(/*is_store=*/true);
 671   case vmIntrinsics::_VectorGatherOp:
 672     return inline_vector_gather_scatter(/*is_scatter*/ false);
 673   case vmIntrinsics::_VectorScatterOp:
 674     return inline_vector_gather_scatter(/*is_scatter*/ true);
 675   case vmIntrinsics::_VectorReductionCoerced:
 676     return inline_vector_reduction();
 677   case vmIntrinsics::_VectorTest:
 678     return inline_vector_test();
 679   case vmIntrinsics::_VectorBlend:
 680     return inline_vector_blend();
 681   case vmIntrinsics::_VectorRearrange:
 682     return inline_vector_rearrange();
 683   case vmIntrinsics::_VectorCompare:
 684     return inline_vector_compare();
 685   case vmIntrinsics::_VectorBroadcastInt:
 686     return inline_vector_broadcast_int();
 687   case vmIntrinsics::_VectorConvert:
 688     return inline_vector_convert();
 689   case vmIntrinsics::_VectorInsert:
 690     return inline_vector_insert();
 691   case vmIntrinsics::_VectorExtract:
 692     return inline_vector_extract();
 693 
 694   case vmIntrinsics::_getObjectSize:
 695     return inline_getObjectSize();
 696 
 697   case vmIntrinsics::_blackhole:
 698     return inline_blackhole();
 699 
 700   default:
 701     // If you get here, it may be that someone has added a new intrinsic
 702     // to the list in vmIntrinsics.hpp without implementing it here.
 703 #ifndef PRODUCT
 704     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 705       tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
 706                     vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
 707     }
 708 #endif
 709     return false;
 710   }
 711 }
 712 
 713 Node* LibraryCallKit::try_to_predicate(int predicate) {
 714   if (!jvms()->has_method()) {
 715     // Root JVMState has a null method.
 716     assert(map()->memory()->Opcode() == Op_Parm, "");
 717     // Insert the memory aliasing node
 718     set_all_memory(reset_memory());
 719   }
 720   assert(merged_memory(), "");
 721 
 722   switch (intrinsic_id()) {
 723   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
 724     return inline_cipherBlockChaining_AESCrypt_predicate(false);
 725   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
 726     return inline_cipherBlockChaining_AESCrypt_predicate(true);
 727   case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
 728     return inline_electronicCodeBook_AESCrypt_predicate(false);
 729   case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
 730     return inline_electronicCodeBook_AESCrypt_predicate(true);
 731   case vmIntrinsics::_counterMode_AESCrypt:
 732     return inline_counterMode_AESCrypt_predicate();
 733   case vmIntrinsics::_digestBase_implCompressMB:
 734     return inline_digestBase_implCompressMB_predicate(predicate);
 735   case vmIntrinsics::_galoisCounterMode_AESCrypt:
 736     return inline_galoisCounterMode_AESCrypt_predicate();
 737 
 738   default:
 739     // If you get here, it may be that someone has added a new intrinsic
 740     // to the list in vmIntrinsics.hpp without implementing it here.
 741 #ifndef PRODUCT
 742     if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
 743       tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
 744                     vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
 745     }
 746 #endif
 747     Node* slow_ctl = control();
 748     set_control(top()); // No fast path instrinsic
 749     return slow_ctl;
 750   }
 751 }
 752 
 753 //------------------------------set_result-------------------------------
 754 // Helper function for finishing intrinsics.
 755 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
 756   record_for_igvn(region);
 757   set_control(_gvn.transform(region));
 758   set_result( _gvn.transform(value));
 759   assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
 760 }
 761 
 762 //------------------------------generate_guard---------------------------
 763 // Helper function for generating guarded fast-slow graph structures.
 764 // The given 'test', if true, guards a slow path.  If the test fails
 765 // then a fast path can be taken.  (We generally hope it fails.)
 766 // In all cases, GraphKit::control() is updated to the fast path.
 767 // The returned value represents the control for the slow path.
 768 // The return value is never 'top'; it is either a valid control
 769 // or NULL if it is obvious that the slow path can never be taken.
 770 // Also, if region and the slow control are not NULL, the slow edge
 771 // is appended to the region.
 772 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
 773   if (stopped()) {
 774     // Already short circuited.
 775     return NULL;
 776   }
 777 
 778   // Build an if node and its projections.
 779   // If test is true we take the slow path, which we assume is uncommon.
 780   if (_gvn.type(test) == TypeInt::ZERO) {
 781     // The slow branch is never taken.  No need to build this guard.
 782     return NULL;
 783   }
 784 
 785   IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
 786 
 787   Node* if_slow = _gvn.transform(new IfTrueNode(iff));
 788   if (if_slow == top()) {
 789     // The slow branch is never taken.  No need to build this guard.
 790     return NULL;
 791   }
 792 
 793   if (region != NULL)
 794     region->add_req(if_slow);
 795 
 796   Node* if_fast = _gvn.transform(new IfFalseNode(iff));
 797   set_control(if_fast);
 798 
 799   return if_slow;
 800 }
 801 
 802 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
 803   return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
 804 }
 805 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
 806   return generate_guard(test, region, PROB_FAIR);
 807 }
 808 
 809 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
 810                                                      Node* *pos_index) {
 811   if (stopped())
 812     return NULL;                // already stopped
 813   if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
 814     return NULL;                // index is already adequately typed
 815   Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
 816   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
 817   Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
 818   if (is_neg != NULL && pos_index != NULL) {
 819     // Emulate effect of Parse::adjust_map_after_if.
 820     Node* ccast = new CastIINode(index, TypeInt::POS);
 821     ccast->set_req(0, control());
 822     (*pos_index) = _gvn.transform(ccast);
 823   }
 824   return is_neg;
 825 }
 826 
 827 // Make sure that 'position' is a valid limit index, in [0..length].
 828 // There are two equivalent plans for checking this:
 829 //   A. (offset + copyLength)  unsigned<=  arrayLength
 830 //   B. offset  <=  (arrayLength - copyLength)
 831 // We require that all of the values above, except for the sum and
 832 // difference, are already known to be non-negative.
 833 // Plan A is robust in the face of overflow, if offset and copyLength
 834 // are both hugely positive.
 835 //
 836 // Plan B is less direct and intuitive, but it does not overflow at
 837 // all, since the difference of two non-negatives is always
 838 // representable.  Whenever Java methods must perform the equivalent
 839 // check they generally use Plan B instead of Plan A.
 840 // For the moment we use Plan A.
 841 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
 842                                                   Node* subseq_length,
 843                                                   Node* array_length,
 844                                                   RegionNode* region) {
 845   if (stopped())
 846     return NULL;                // already stopped
 847   bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
 848   if (zero_offset && subseq_length->eqv_uncast(array_length))
 849     return NULL;                // common case of whole-array copy
 850   Node* last = subseq_length;
 851   if (!zero_offset)             // last += offset
 852     last = _gvn.transform(new AddINode(last, offset));
 853   Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
 854   Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
 855   Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
 856   return is_over;
 857 }
 858 
 859 // Emit range checks for the given String.value byte array
 860 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
 861   if (stopped()) {
 862     return; // already stopped
 863   }
 864   RegionNode* bailout = new RegionNode(1);
 865   record_for_igvn(bailout);
 866   if (char_count) {
 867     // Convert char count to byte count
 868     count = _gvn.transform(new LShiftINode(count, intcon(1)));
 869   }
 870 
 871   // Offset and count must not be negative
 872   generate_negative_guard(offset, bailout);
 873   generate_negative_guard(count, bailout);
 874   // Offset + count must not exceed length of array
 875   generate_limit_guard(offset, count, load_array_length(array), bailout);
 876 
 877   if (bailout->req() > 1) {
 878     PreserveJVMState pjvms(this);
 879     set_control(_gvn.transform(bailout));
 880     uncommon_trap(Deoptimization::Reason_intrinsic,
 881                   Deoptimization::Action_maybe_recompile);
 882   }
 883 }
 884 
 885 //--------------------------generate_current_thread--------------------
 886 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
 887   ciKlass*    thread_klass = env()->Thread_klass();
 888   const Type* thread_type  = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
 889   Node* thread = _gvn.transform(new ThreadLocalNode());
 890   Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
 891   tls_output = thread;
 892   Node* thread_obj_handle = LoadNode::make(_gvn, NULL, immutable_memory(), p, p->bottom_type()->is_ptr(), TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
 893   thread_obj_handle = _gvn.transform(thread_obj_handle);
 894   return access_load(thread_obj_handle, thread_type, T_OBJECT, IN_NATIVE | C2_IMMUTABLE_MEMORY);
 895 }
 896 
 897 
 898 //------------------------------make_string_method_node------------------------
 899 // Helper method for String intrinsic functions. This version is called with
 900 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
 901 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
 902 // containing the lengths of str1 and str2.
 903 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
 904   Node* result = NULL;
 905   switch (opcode) {
 906   case Op_StrIndexOf:
 907     result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
 908                                 str1_start, cnt1, str2_start, cnt2, ae);
 909     break;
 910   case Op_StrComp:
 911     result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
 912                              str1_start, cnt1, str2_start, cnt2, ae);
 913     break;
 914   case Op_StrEquals:
 915     // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
 916     // Use the constant length if there is one because optimized match rule may exist.
 917     result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
 918                                str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
 919     break;
 920   default:
 921     ShouldNotReachHere();
 922     return NULL;
 923   }
 924 
 925   // All these intrinsics have checks.
 926   C->set_has_split_ifs(true); // Has chance for split-if optimization
 927   clear_upper_avx();
 928 
 929   return _gvn.transform(result);
 930 }
 931 
 932 //------------------------------inline_string_compareTo------------------------
 933 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
 934   Node* arg1 = argument(0);
 935   Node* arg2 = argument(1);
 936 
 937   arg1 = must_be_not_null(arg1, true);
 938   arg2 = must_be_not_null(arg2, true);
 939 
 940   // Get start addr and length of first argument
 941   Node* arg1_start  = array_element_address(arg1, intcon(0), T_BYTE);
 942   Node* arg1_cnt    = load_array_length(arg1);
 943 
 944   // Get start addr and length of second argument
 945   Node* arg2_start  = array_element_address(arg2, intcon(0), T_BYTE);
 946   Node* arg2_cnt    = load_array_length(arg2);
 947 
 948   Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
 949   set_result(result);
 950   return true;
 951 }
 952 
 953 //------------------------------inline_string_equals------------------------
 954 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
 955   Node* arg1 = argument(0);
 956   Node* arg2 = argument(1);
 957 
 958   // paths (plus control) merge
 959   RegionNode* region = new RegionNode(3);
 960   Node* phi = new PhiNode(region, TypeInt::BOOL);
 961 
 962   if (!stopped()) {
 963 
 964     arg1 = must_be_not_null(arg1, true);
 965     arg2 = must_be_not_null(arg2, true);
 966 
 967     // Get start addr and length of first argument
 968     Node* arg1_start  = array_element_address(arg1, intcon(0), T_BYTE);
 969     Node* arg1_cnt    = load_array_length(arg1);
 970 
 971     // Get start addr and length of second argument
 972     Node* arg2_start  = array_element_address(arg2, intcon(0), T_BYTE);
 973     Node* arg2_cnt    = load_array_length(arg2);
 974 
 975     // Check for arg1_cnt != arg2_cnt
 976     Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
 977     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
 978     Node* if_ne = generate_slow_guard(bol, NULL);
 979     if (if_ne != NULL) {
 980       phi->init_req(2, intcon(0));
 981       region->init_req(2, if_ne);
 982     }
 983 
 984     // Check for count == 0 is done by assembler code for StrEquals.
 985 
 986     if (!stopped()) {
 987       Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
 988       phi->init_req(1, equals);
 989       region->init_req(1, control());
 990     }
 991   }
 992 
 993   // post merge
 994   set_control(_gvn.transform(region));
 995   record_for_igvn(region);
 996 
 997   set_result(_gvn.transform(phi));
 998   return true;
 999 }
1000 
1001 //------------------------------inline_array_equals----------------------------
1002 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1003   assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1004   Node* arg1 = argument(0);
1005   Node* arg2 = argument(1);
1006 
1007   const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1008   set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1009   clear_upper_avx();
1010 
1011   return true;
1012 }
1013 
1014 //------------------------------inline_hasNegatives------------------------------
1015 bool LibraryCallKit::inline_hasNegatives() {
1016   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1017     return false;
1018   }
1019 
1020   assert(callee()->signature()->size() == 3, "hasNegatives has 3 parameters");
1021   // no receiver since it is static method
1022   Node* ba         = argument(0);
1023   Node* offset     = argument(1);
1024   Node* len        = argument(2);
1025 
1026   ba = must_be_not_null(ba, true);
1027 
1028   // Range checks
1029   generate_string_range_check(ba, offset, len, false);
1030   if (stopped()) {
1031     return true;
1032   }
1033   Node* ba_start = array_element_address(ba, offset, T_BYTE);
1034   Node* result = new HasNegativesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1035   set_result(_gvn.transform(result));
1036   return true;
1037 }
1038 
1039 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1040   Node* index = argument(0);
1041   Node* length = bt == T_INT ? argument(1) : argument(2);
1042   if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1043     return false;
1044   }
1045 
1046   // check that length is positive
1047   Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1048   Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1049 
1050   {
1051     BuildCutout unless(this, len_pos_bol, PROB_MAX);
1052     uncommon_trap(Deoptimization::Reason_intrinsic,
1053                   Deoptimization::Action_make_not_entrant);
1054   }
1055 
1056   if (stopped()) {
1057     // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1058     return true;
1059   }
1060 
1061   // length is now known postive, add a cast node to make this explicit
1062   jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1063   Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1064   casted_length = _gvn.transform(casted_length);
1065   replace_in_map(length, casted_length);
1066   length = casted_length;
1067 
1068   // Use an unsigned comparison for the range check itself
1069   Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1070   BoolTest::mask btest = BoolTest::lt;
1071   Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1072   RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1073   _gvn.set_type(rc, rc->Value(&_gvn));
1074   if (!rc_bool->is_Con()) {
1075     record_for_igvn(rc);
1076   }
1077   set_control(_gvn.transform(new IfTrueNode(rc)));
1078   {
1079     PreserveJVMState pjvms(this);
1080     set_control(_gvn.transform(new IfFalseNode(rc)));
1081     uncommon_trap(Deoptimization::Reason_range_check,
1082                   Deoptimization::Action_make_not_entrant);
1083   }
1084 
1085   if (stopped()) {
1086     // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1087     return true;
1088   }
1089 
1090   // index is now known to be >= 0 and < length, cast it
1091   Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), bt);
1092   result = _gvn.transform(result);
1093   set_result(result);
1094   replace_in_map(index, result);
1095   clear_upper_avx();
1096   return true;
1097 }
1098 
1099 //------------------------------inline_string_indexOf------------------------
1100 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1101   if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1102     return false;
1103   }
1104   Node* src = argument(0);
1105   Node* tgt = argument(1);
1106 
1107   // Make the merge point
1108   RegionNode* result_rgn = new RegionNode(4);
1109   Node*       result_phi = new PhiNode(result_rgn, TypeInt::INT);
1110 
1111   src = must_be_not_null(src, true);
1112   tgt = must_be_not_null(tgt, true);
1113 
1114   // Get start addr and length of source string
1115   Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1116   Node* src_count = load_array_length(src);
1117 
1118   // Get start addr and length of substring
1119   Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1120   Node* tgt_count = load_array_length(tgt);
1121 
1122   if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1123     // Divide src size by 2 if String is UTF16 encoded
1124     src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1125   }
1126   if (ae == StrIntrinsicNode::UU) {
1127     // Divide substring size by 2 if String is UTF16 encoded
1128     tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1129   }
1130 
1131   Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1132   if (result != NULL) {
1133     result_phi->init_req(3, result);
1134     result_rgn->init_req(3, control());
1135   }
1136   set_control(_gvn.transform(result_rgn));
1137   record_for_igvn(result_rgn);
1138   set_result(_gvn.transform(result_phi));
1139 
1140   return true;
1141 }
1142 
1143 //-----------------------------inline_string_indexOf-----------------------
1144 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1145   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1146     return false;
1147   }
1148   if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1149     return false;
1150   }
1151   assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1152   Node* src         = argument(0); // byte[]
1153   Node* src_count   = argument(1); // char count
1154   Node* tgt         = argument(2); // byte[]
1155   Node* tgt_count   = argument(3); // char count
1156   Node* from_index  = argument(4); // char index
1157 
1158   src = must_be_not_null(src, true);
1159   tgt = must_be_not_null(tgt, true);
1160 
1161   // Multiply byte array index by 2 if String is UTF16 encoded
1162   Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1163   src_count = _gvn.transform(new SubINode(src_count, from_index));
1164   Node* src_start = array_element_address(src, src_offset, T_BYTE);
1165   Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1166 
1167   // Range checks
1168   generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1169   generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1170   if (stopped()) {
1171     return true;
1172   }
1173 
1174   RegionNode* region = new RegionNode(5);
1175   Node* phi = new PhiNode(region, TypeInt::INT);
1176 
1177   Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1178   if (result != NULL) {
1179     // The result is index relative to from_index if substring was found, -1 otherwise.
1180     // Generate code which will fold into cmove.
1181     Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1182     Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1183 
1184     Node* if_lt = generate_slow_guard(bol, NULL);
1185     if (if_lt != NULL) {
1186       // result == -1
1187       phi->init_req(3, result);
1188       region->init_req(3, if_lt);
1189     }
1190     if (!stopped()) {
1191       result = _gvn.transform(new AddINode(result, from_index));
1192       phi->init_req(4, result);
1193       region->init_req(4, control());
1194     }
1195   }
1196 
1197   set_control(_gvn.transform(region));
1198   record_for_igvn(region);
1199   set_result(_gvn.transform(phi));
1200   clear_upper_avx();
1201 
1202   return true;
1203 }
1204 
1205 // Create StrIndexOfNode with fast path checks
1206 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1207                                         RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1208   // Check for substr count > string count
1209   Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1210   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1211   Node* if_gt = generate_slow_guard(bol, NULL);
1212   if (if_gt != NULL) {
1213     phi->init_req(1, intcon(-1));
1214     region->init_req(1, if_gt);
1215   }
1216   if (!stopped()) {
1217     // Check for substr count == 0
1218     cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1219     bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1220     Node* if_zero = generate_slow_guard(bol, NULL);
1221     if (if_zero != NULL) {
1222       phi->init_req(2, intcon(0));
1223       region->init_req(2, if_zero);
1224     }
1225   }
1226   if (!stopped()) {
1227     return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1228   }
1229   return NULL;
1230 }
1231 
1232 //-----------------------------inline_string_indexOfChar-----------------------
1233 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1234   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1235     return false;
1236   }
1237   if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1238     return false;
1239   }
1240   assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1241   Node* src         = argument(0); // byte[]
1242   Node* tgt         = argument(1); // tgt is int ch
1243   Node* from_index  = argument(2);
1244   Node* max         = argument(3);
1245 
1246   src = must_be_not_null(src, true);
1247 
1248   Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1249   Node* src_start = array_element_address(src, src_offset, T_BYTE);
1250   Node* src_count = _gvn.transform(new SubINode(max, from_index));
1251 
1252   // Range checks
1253   generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1254   if (stopped()) {
1255     return true;
1256   }
1257 
1258   RegionNode* region = new RegionNode(3);
1259   Node* phi = new PhiNode(region, TypeInt::INT);
1260 
1261   Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, tgt, ae);
1262   C->set_has_split_ifs(true); // Has chance for split-if optimization
1263   _gvn.transform(result);
1264 
1265   Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1266   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1267 
1268   Node* if_lt = generate_slow_guard(bol, NULL);
1269   if (if_lt != NULL) {
1270     // result == -1
1271     phi->init_req(2, result);
1272     region->init_req(2, if_lt);
1273   }
1274   if (!stopped()) {
1275     result = _gvn.transform(new AddINode(result, from_index));
1276     phi->init_req(1, result);
1277     region->init_req(1, control());
1278   }
1279   set_control(_gvn.transform(region));
1280   record_for_igvn(region);
1281   set_result(_gvn.transform(phi));
1282 
1283   return true;
1284 }
1285 //---------------------------inline_string_copy---------------------
1286 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1287 //   int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1288 //   int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1289 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1290 //   void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1291 //   void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1292 bool LibraryCallKit::inline_string_copy(bool compress) {
1293   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1294     return false;
1295   }
1296   int nargs = 5;  // 2 oops, 3 ints
1297   assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1298 
1299   Node* src         = argument(0);
1300   Node* src_offset  = argument(1);
1301   Node* dst         = argument(2);
1302   Node* dst_offset  = argument(3);
1303   Node* length      = argument(4);
1304 
1305   // Check for allocation before we add nodes that would confuse
1306   // tightly_coupled_allocation()
1307   AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1308 
1309   // Figure out the size and type of the elements we will be copying.
1310   const Type* src_type = src->Value(&_gvn);
1311   const Type* dst_type = dst->Value(&_gvn);
1312   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1313   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
1314   assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1315          (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1316          "Unsupported array types for inline_string_copy");
1317 
1318   src = must_be_not_null(src, true);
1319   dst = must_be_not_null(dst, true);
1320 
1321   // Convert char[] offsets to byte[] offsets
1322   bool convert_src = (compress && src_elem == T_BYTE);
1323   bool convert_dst = (!compress && dst_elem == T_BYTE);
1324   if (convert_src) {
1325     src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1326   } else if (convert_dst) {
1327     dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1328   }
1329 
1330   // Range checks
1331   generate_string_range_check(src, src_offset, length, convert_src);
1332   generate_string_range_check(dst, dst_offset, length, convert_dst);
1333   if (stopped()) {
1334     return true;
1335   }
1336 
1337   Node* src_start = array_element_address(src, src_offset, src_elem);
1338   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1339   // 'src_start' points to src array + scaled offset
1340   // 'dst_start' points to dst array + scaled offset
1341   Node* count = NULL;
1342   if (compress) {
1343     count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1344   } else {
1345     inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1346   }
1347 
1348   if (alloc != NULL) {
1349     if (alloc->maybe_set_complete(&_gvn)) {
1350       // "You break it, you buy it."
1351       InitializeNode* init = alloc->initialization();
1352       assert(init->is_complete(), "we just did this");
1353       init->set_complete_with_arraycopy();
1354       assert(dst->is_CheckCastPP(), "sanity");
1355       assert(dst->in(0)->in(0) == init, "dest pinned");
1356     }
1357     // Do not let stores that initialize this object be reordered with
1358     // a subsequent store that would make this object accessible by
1359     // other threads.
1360     // Record what AllocateNode this StoreStore protects so that
1361     // escape analysis can go from the MemBarStoreStoreNode to the
1362     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1363     // based on the escape status of the AllocateNode.
1364     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1365   }
1366   if (compress) {
1367     set_result(_gvn.transform(count));
1368   }
1369   clear_upper_avx();
1370 
1371   return true;
1372 }
1373 
1374 #ifdef _LP64
1375 #define XTOP ,top() /*additional argument*/
1376 #else  //_LP64
1377 #define XTOP        /*no additional argument*/
1378 #endif //_LP64
1379 
1380 //------------------------inline_string_toBytesU--------------------------
1381 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1382 bool LibraryCallKit::inline_string_toBytesU() {
1383   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1384     return false;
1385   }
1386   // Get the arguments.
1387   Node* value     = argument(0);
1388   Node* offset    = argument(1);
1389   Node* length    = argument(2);
1390 
1391   Node* newcopy = NULL;
1392 
1393   // Set the original stack and the reexecute bit for the interpreter to reexecute
1394   // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1395   { PreserveReexecuteState preexecs(this);
1396     jvms()->set_should_reexecute(true);
1397 
1398     // Check if a null path was taken unconditionally.
1399     value = null_check(value);
1400 
1401     RegionNode* bailout = new RegionNode(1);
1402     record_for_igvn(bailout);
1403 
1404     // Range checks
1405     generate_negative_guard(offset, bailout);
1406     generate_negative_guard(length, bailout);
1407     generate_limit_guard(offset, length, load_array_length(value), bailout);
1408     // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1409     generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1410 
1411     if (bailout->req() > 1) {
1412       PreserveJVMState pjvms(this);
1413       set_control(_gvn.transform(bailout));
1414       uncommon_trap(Deoptimization::Reason_intrinsic,
1415                     Deoptimization::Action_maybe_recompile);
1416     }
1417     if (stopped()) {
1418       return true;
1419     }
1420 
1421     Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1422     Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1423     newcopy = new_array(klass_node, size, 0);  // no arguments to push
1424     AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1425     guarantee(alloc != NULL, "created above");
1426 
1427     // Calculate starting addresses.
1428     Node* src_start = array_element_address(value, offset, T_CHAR);
1429     Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1430 
1431     // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1432     const TypeInt* toffset = gvn().type(offset)->is_int();
1433     bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1434 
1435     // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1436     const char* copyfunc_name = "arraycopy";
1437     address     copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1438     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1439                       OptoRuntime::fast_arraycopy_Type(),
1440                       copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1441                       src_start, dst_start, ConvI2X(length) XTOP);
1442     // Do not let reads from the cloned object float above the arraycopy.
1443     if (alloc->maybe_set_complete(&_gvn)) {
1444       // "You break it, you buy it."
1445       InitializeNode* init = alloc->initialization();
1446       assert(init->is_complete(), "we just did this");
1447       init->set_complete_with_arraycopy();
1448       assert(newcopy->is_CheckCastPP(), "sanity");
1449       assert(newcopy->in(0)->in(0) == init, "dest pinned");
1450     }
1451     // Do not let stores that initialize this object be reordered with
1452     // a subsequent store that would make this object accessible by
1453     // other threads.
1454     // Record what AllocateNode this StoreStore protects so that
1455     // escape analysis can go from the MemBarStoreStoreNode to the
1456     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1457     // based on the escape status of the AllocateNode.
1458     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1459   } // original reexecute is set back here
1460 
1461   C->set_has_split_ifs(true); // Has chance for split-if optimization
1462   if (!stopped()) {
1463     set_result(newcopy);
1464   }
1465   clear_upper_avx();
1466 
1467   return true;
1468 }
1469 
1470 //------------------------inline_string_getCharsU--------------------------
1471 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1472 bool LibraryCallKit::inline_string_getCharsU() {
1473   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1474     return false;
1475   }
1476 
1477   // Get the arguments.
1478   Node* src       = argument(0);
1479   Node* src_begin = argument(1);
1480   Node* src_end   = argument(2); // exclusive offset (i < src_end)
1481   Node* dst       = argument(3);
1482   Node* dst_begin = argument(4);
1483 
1484   // Check for allocation before we add nodes that would confuse
1485   // tightly_coupled_allocation()
1486   AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1487 
1488   // Check if a null path was taken unconditionally.
1489   src = null_check(src);
1490   dst = null_check(dst);
1491   if (stopped()) {
1492     return true;
1493   }
1494 
1495   // Get length and convert char[] offset to byte[] offset
1496   Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1497   src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1498 
1499   // Range checks
1500   generate_string_range_check(src, src_begin, length, true);
1501   generate_string_range_check(dst, dst_begin, length, false);
1502   if (stopped()) {
1503     return true;
1504   }
1505 
1506   if (!stopped()) {
1507     // Calculate starting addresses.
1508     Node* src_start = array_element_address(src, src_begin, T_BYTE);
1509     Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1510 
1511     // Check if array addresses are aligned to HeapWordSize
1512     const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1513     const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1514     bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1515                    tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1516 
1517     // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1518     const char* copyfunc_name = "arraycopy";
1519     address     copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1520     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1521                       OptoRuntime::fast_arraycopy_Type(),
1522                       copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1523                       src_start, dst_start, ConvI2X(length) XTOP);
1524     // Do not let reads from the cloned object float above the arraycopy.
1525     if (alloc != NULL) {
1526       if (alloc->maybe_set_complete(&_gvn)) {
1527         // "You break it, you buy it."
1528         InitializeNode* init = alloc->initialization();
1529         assert(init->is_complete(), "we just did this");
1530         init->set_complete_with_arraycopy();
1531         assert(dst->is_CheckCastPP(), "sanity");
1532         assert(dst->in(0)->in(0) == init, "dest pinned");
1533       }
1534       // Do not let stores that initialize this object be reordered with
1535       // a subsequent store that would make this object accessible by
1536       // other threads.
1537       // Record what AllocateNode this StoreStore protects so that
1538       // escape analysis can go from the MemBarStoreStoreNode to the
1539       // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1540       // based on the escape status of the AllocateNode.
1541       insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1542     } else {
1543       insert_mem_bar(Op_MemBarCPUOrder);
1544     }
1545   }
1546 
1547   C->set_has_split_ifs(true); // Has chance for split-if optimization
1548   return true;
1549 }
1550 
1551 //----------------------inline_string_char_access----------------------------
1552 // Store/Load char to/from byte[] array.
1553 // static void StringUTF16.putChar(byte[] val, int index, int c)
1554 // static char StringUTF16.getChar(byte[] val, int index)
1555 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1556   Node* value  = argument(0);
1557   Node* index  = argument(1);
1558   Node* ch = is_store ? argument(2) : NULL;
1559 
1560   // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1561   // correctly requires matched array shapes.
1562   assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1563           "sanity: byte[] and char[] bases agree");
1564   assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1565           "sanity: byte[] and char[] scales agree");
1566 
1567   // Bail when getChar over constants is requested: constant folding would
1568   // reject folding mismatched char access over byte[]. A normal inlining for getChar
1569   // Java method would constant fold nicely instead.
1570   if (!is_store && value->is_Con() && index->is_Con()) {
1571     return false;
1572   }
1573 
1574   value = must_be_not_null(value, true);
1575 
1576   Node* adr = array_element_address(value, index, T_CHAR);
1577   if (adr->is_top()) {
1578     return false;
1579   }
1580   if (is_store) {
1581     access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1582   } else {
1583     ch = access_load_at(value, adr, TypeAryPtr::BYTES, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
1584     set_result(ch);
1585   }
1586   return true;
1587 }
1588 
1589 //--------------------------round_double_node--------------------------------
1590 // Round a double node if necessary.
1591 Node* LibraryCallKit::round_double_node(Node* n) {
1592   if (Matcher::strict_fp_requires_explicit_rounding) {
1593 #ifdef IA32
1594     if (UseSSE < 2) {
1595       n = _gvn.transform(new RoundDoubleNode(NULL, n));
1596     }
1597 #else
1598     Unimplemented();
1599 #endif // IA32
1600   }
1601   return n;
1602 }
1603 
1604 //------------------------------inline_math-----------------------------------
1605 // public static double Math.abs(double)
1606 // public static double Math.sqrt(double)
1607 // public static double Math.log(double)
1608 // public static double Math.log10(double)
1609 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1610   Node* arg = round_double_node(argument(0));
1611   Node* n = NULL;
1612   switch (id) {
1613   case vmIntrinsics::_dabs:   n = new AbsDNode(                arg);  break;
1614   case vmIntrinsics::_dsqrt:
1615   case vmIntrinsics::_dsqrt_strict:
1616                               n = new SqrtDNode(C, control(),  arg);  break;
1617   case vmIntrinsics::_ceil:   n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1618   case vmIntrinsics::_floor:  n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1619   case vmIntrinsics::_rint:   n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1620   case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1621   case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1622   default:  fatal_unexpected_iid(id);  break;
1623   }
1624   set_result(_gvn.transform(n));
1625   return true;
1626 }
1627 
1628 //------------------------------inline_math-----------------------------------
1629 // public static float Math.abs(float)
1630 // public static int Math.abs(int)
1631 // public static long Math.abs(long)
1632 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1633   Node* arg = argument(0);
1634   Node* n = NULL;
1635   switch (id) {
1636   case vmIntrinsics::_fabs:   n = new AbsFNode(                arg);  break;
1637   case vmIntrinsics::_iabs:   n = new AbsINode(                arg);  break;
1638   case vmIntrinsics::_labs:   n = new AbsLNode(                arg);  break;
1639   case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1640   case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1641   default:  fatal_unexpected_iid(id);  break;
1642   }
1643   set_result(_gvn.transform(n));
1644   return true;
1645 }
1646 
1647 //------------------------------runtime_math-----------------------------
1648 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1649   assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1650          "must be (DD)D or (D)D type");
1651 
1652   // Inputs
1653   Node* a = round_double_node(argument(0));
1654   Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
1655 
1656   const TypePtr* no_memory_effects = NULL;
1657   Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1658                                  no_memory_effects,
1659                                  a, top(), b, b ? top() : NULL);
1660   Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1661 #ifdef ASSERT
1662   Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1663   assert(value_top == top(), "second value must be top");
1664 #endif
1665 
1666   set_result(value);
1667   return true;
1668 }
1669 
1670 //------------------------------inline_math_pow-----------------------------
1671 bool LibraryCallKit::inline_math_pow() {
1672   Node* exp = round_double_node(argument(2));
1673   const TypeD* d = _gvn.type(exp)->isa_double_constant();
1674   if (d != NULL) {
1675     if (d->getd() == 2.0) {
1676       // Special case: pow(x, 2.0) => x * x
1677       Node* base = round_double_node(argument(0));
1678       set_result(_gvn.transform(new MulDNode(base, base)));
1679       return true;
1680     } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1681       // Special case: pow(x, 0.5) => sqrt(x)
1682       Node* base = round_double_node(argument(0));
1683       Node* zero = _gvn.zerocon(T_DOUBLE);
1684 
1685       RegionNode* region = new RegionNode(3);
1686       Node* phi = new PhiNode(region, Type::DOUBLE);
1687 
1688       Node* cmp  = _gvn.transform(new CmpDNode(base, zero));
1689       // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1690       // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1691       // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1692       Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1693 
1694       Node* if_pow = generate_slow_guard(test, NULL);
1695       Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1696       phi->init_req(1, value_sqrt);
1697       region->init_req(1, control());
1698 
1699       if (if_pow != NULL) {
1700         set_control(if_pow);
1701         address target = StubRoutines::dpow() != NULL ? StubRoutines::dpow() :
1702                                                         CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1703         const TypePtr* no_memory_effects = NULL;
1704         Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1705                                        no_memory_effects, base, top(), exp, top());
1706         Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1707 #ifdef ASSERT
1708         Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1709         assert(value_top == top(), "second value must be top");
1710 #endif
1711         phi->init_req(2, value_pow);
1712         region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1713       }
1714 
1715       C->set_has_split_ifs(true); // Has chance for split-if optimization
1716       set_control(_gvn.transform(region));
1717       record_for_igvn(region);
1718       set_result(_gvn.transform(phi));
1719 
1720       return true;
1721     }
1722   }
1723 
1724   return StubRoutines::dpow() != NULL ?
1725     runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(),  "dpow") :
1726     runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow),  "POW");
1727 }
1728 
1729 //------------------------------inline_math_native-----------------------------
1730 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1731 #define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
1732   switch (id) {
1733     // These intrinsics are not properly supported on all hardware
1734   case vmIntrinsics::_dsin:
1735     return StubRoutines::dsin() != NULL ?
1736       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1737       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin),   "SIN");
1738   case vmIntrinsics::_dcos:
1739     return StubRoutines::dcos() != NULL ?
1740       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1741       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos),   "COS");
1742   case vmIntrinsics::_dtan:
1743     return StubRoutines::dtan() != NULL ?
1744       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1745       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan), "TAN");
1746   case vmIntrinsics::_dlog:
1747     return StubRoutines::dlog() != NULL ?
1748       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1749       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog),   "LOG");
1750   case vmIntrinsics::_dlog10:
1751     return StubRoutines::dlog10() != NULL ?
1752       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1753       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
1754 
1755     // These intrinsics are supported on all hardware
1756   case vmIntrinsics::_ceil:
1757   case vmIntrinsics::_floor:
1758   case vmIntrinsics::_rint:   return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1759   case vmIntrinsics::_dsqrt:
1760   case vmIntrinsics::_dsqrt_strict:
1761                               return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1762   case vmIntrinsics::_dabs:   return Matcher::has_match_rule(Op_AbsD)   ? inline_double_math(id) : false;
1763   case vmIntrinsics::_fabs:   return Matcher::match_rule_supported(Op_AbsF)   ? inline_math(id) : false;
1764   case vmIntrinsics::_iabs:   return Matcher::match_rule_supported(Op_AbsI)   ? inline_math(id) : false;
1765   case vmIntrinsics::_labs:   return Matcher::match_rule_supported(Op_AbsL)   ? inline_math(id) : false;
1766 
1767   case vmIntrinsics::_dexp:
1768     return StubRoutines::dexp() != NULL ?
1769       runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(),  "dexp") :
1770       runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dexp),  "EXP");
1771 #undef FN_PTR
1772 
1773   case vmIntrinsics::_dpow:      return inline_math_pow();
1774   case vmIntrinsics::_dcopySign: return inline_double_math(id);
1775   case vmIntrinsics::_fcopySign: return inline_math(id);
1776   case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1777   case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1778 
1779    // These intrinsics are not yet correctly implemented
1780   case vmIntrinsics::_datan2:
1781     return false;
1782 
1783   default:
1784     fatal_unexpected_iid(id);
1785     return false;
1786   }
1787 }
1788 
1789 static bool is_simple_name(Node* n) {
1790   return (n->req() == 1         // constant
1791           || (n->is_Type() && n->as_Type()->type()->singleton())
1792           || n->is_Proj()       // parameter or return value
1793           || n->is_Phi()        // local of some sort
1794           );
1795 }
1796 
1797 //----------------------------inline_notify-----------------------------------*
1798 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1799   const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1800   address func;
1801   if (id == vmIntrinsics::_notify) {
1802     func = OptoRuntime::monitor_notify_Java();
1803   } else {
1804     func = OptoRuntime::monitor_notifyAll_Java();
1805   }
1806   Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, NULL, TypeRawPtr::BOTTOM, argument(0));
1807   make_slow_call_ex(call, env()->Throwable_klass(), false);
1808   return true;
1809 }
1810 
1811 
1812 //----------------------------inline_min_max-----------------------------------
1813 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1814   set_result(generate_min_max(id, argument(0), argument(1)));
1815   return true;
1816 }
1817 
1818 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1819   Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1820   IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1821   Node* fast_path = _gvn.transform( new IfFalseNode(check));
1822   Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1823 
1824   {
1825     PreserveJVMState pjvms(this);
1826     PreserveReexecuteState preexecs(this);
1827     jvms()->set_should_reexecute(true);
1828 
1829     set_control(slow_path);
1830     set_i_o(i_o());
1831 
1832     uncommon_trap(Deoptimization::Reason_intrinsic,
1833                   Deoptimization::Action_none);
1834   }
1835 
1836   set_control(fast_path);
1837   set_result(math);
1838 }
1839 
1840 template <typename OverflowOp>
1841 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1842   typedef typename OverflowOp::MathOp MathOp;
1843 
1844   MathOp* mathOp = new MathOp(arg1, arg2);
1845   Node* operation = _gvn.transform( mathOp );
1846   Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1847   inline_math_mathExact(operation, ofcheck);
1848   return true;
1849 }
1850 
1851 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1852   return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1853 }
1854 
1855 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1856   return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1857 }
1858 
1859 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1860   return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1861 }
1862 
1863 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1864   return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1865 }
1866 
1867 bool LibraryCallKit::inline_math_negateExactI() {
1868   return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1869 }
1870 
1871 bool LibraryCallKit::inline_math_negateExactL() {
1872   return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1873 }
1874 
1875 bool LibraryCallKit::inline_math_multiplyExactI() {
1876   return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1877 }
1878 
1879 bool LibraryCallKit::inline_math_multiplyExactL() {
1880   return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1881 }
1882 
1883 bool LibraryCallKit::inline_math_multiplyHigh() {
1884   set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1885   return true;
1886 }
1887 
1888 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1889   set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1890   return true;
1891 }
1892 
1893 Node*
1894 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1895   // These are the candidate return value:
1896   Node* xvalue = x0;
1897   Node* yvalue = y0;
1898 
1899   if (xvalue == yvalue) {
1900     return xvalue;
1901   }
1902 
1903   bool want_max = (id == vmIntrinsics::_max || id == vmIntrinsics::_max_strict);
1904 
1905   const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
1906   const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
1907   if (txvalue == NULL || tyvalue == NULL)  return top();
1908   // This is not really necessary, but it is consistent with a
1909   // hypothetical MaxINode::Value method:
1910   int widen = MAX2(txvalue->_widen, tyvalue->_widen);
1911 
1912   // %%% This folding logic should (ideally) be in a different place.
1913   // Some should be inside IfNode, and there to be a more reliable
1914   // transformation of ?: style patterns into cmoves.  We also want
1915   // more powerful optimizations around cmove and min/max.
1916 
1917   // Try to find a dominating comparison of these guys.
1918   // It can simplify the index computation for Arrays.copyOf
1919   // and similar uses of System.arraycopy.
1920   // First, compute the normalized version of CmpI(x, y).
1921   int   cmp_op = Op_CmpI;
1922   Node* xkey = xvalue;
1923   Node* ykey = yvalue;
1924   Node* ideal_cmpxy = _gvn.transform(new CmpINode(xkey, ykey));
1925   if (ideal_cmpxy->is_Cmp()) {
1926     // E.g., if we have CmpI(length - offset, count),
1927     // it might idealize to CmpI(length, count + offset)
1928     cmp_op = ideal_cmpxy->Opcode();
1929     xkey = ideal_cmpxy->in(1);
1930     ykey = ideal_cmpxy->in(2);
1931   }
1932 
1933   // Start by locating any relevant comparisons.
1934   Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
1935   Node* cmpxy = NULL;
1936   Node* cmpyx = NULL;
1937   for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
1938     Node* cmp = start_from->fast_out(k);
1939     if (cmp->outcnt() > 0 &&            // must have prior uses
1940         cmp->in(0) == NULL &&           // must be context-independent
1941         cmp->Opcode() == cmp_op) {      // right kind of compare
1942       if (cmp->in(1) == xkey && cmp->in(2) == ykey)  cmpxy = cmp;
1943       if (cmp->in(1) == ykey && cmp->in(2) == xkey)  cmpyx = cmp;
1944     }
1945   }
1946 
1947   const int NCMPS = 2;
1948   Node* cmps[NCMPS] = { cmpxy, cmpyx };
1949   int cmpn;
1950   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1951     if (cmps[cmpn] != NULL)  break;     // find a result
1952   }
1953   if (cmpn < NCMPS) {
1954     // Look for a dominating test that tells us the min and max.
1955     int depth = 0;                // Limit search depth for speed
1956     Node* dom = control();
1957     for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
1958       if (++depth >= 100)  break;
1959       Node* ifproj = dom;
1960       if (!ifproj->is_Proj())  continue;
1961       Node* iff = ifproj->in(0);
1962       if (!iff->is_If())  continue;
1963       Node* bol = iff->in(1);
1964       if (!bol->is_Bool())  continue;
1965       Node* cmp = bol->in(1);
1966       if (cmp == NULL)  continue;
1967       for (cmpn = 0; cmpn < NCMPS; cmpn++)
1968         if (cmps[cmpn] == cmp)  break;
1969       if (cmpn == NCMPS)  continue;
1970       BoolTest::mask btest = bol->as_Bool()->_test._test;
1971       if (ifproj->is_IfFalse())  btest = BoolTest(btest).negate();
1972       if (cmp->in(1) == ykey)    btest = BoolTest(btest).commute();
1973       // At this point, we know that 'x btest y' is true.
1974       switch (btest) {
1975       case BoolTest::eq:
1976         // They are proven equal, so we can collapse the min/max.
1977         // Either value is the answer.  Choose the simpler.
1978         if (is_simple_name(yvalue) && !is_simple_name(xvalue))
1979           return yvalue;
1980         return xvalue;
1981       case BoolTest::lt:          // x < y
1982       case BoolTest::le:          // x <= y
1983         return (want_max ? yvalue : xvalue);
1984       case BoolTest::gt:          // x > y
1985       case BoolTest::ge:          // x >= y
1986         return (want_max ? xvalue : yvalue);
1987       default:
1988         break;
1989       }
1990     }
1991   }
1992 
1993   // We failed to find a dominating test.
1994   // Let's pick a test that might GVN with prior tests.
1995   Node*          best_bol   = NULL;
1996   BoolTest::mask best_btest = BoolTest::illegal;
1997   for (cmpn = 0; cmpn < NCMPS; cmpn++) {
1998     Node* cmp = cmps[cmpn];
1999     if (cmp == NULL)  continue;
2000     for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
2001       Node* bol = cmp->fast_out(j);
2002       if (!bol->is_Bool())  continue;
2003       BoolTest::mask btest = bol->as_Bool()->_test._test;
2004       if (btest == BoolTest::eq || btest == BoolTest::ne)  continue;
2005       if (cmp->in(1) == ykey)   btest = BoolTest(btest).commute();
2006       if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
2007         best_bol   = bol->as_Bool();
2008         best_btest = btest;
2009       }
2010     }
2011   }
2012 
2013   Node* answer_if_true  = NULL;
2014   Node* answer_if_false = NULL;
2015   switch (best_btest) {
2016   default:
2017     if (cmpxy == NULL)
2018       cmpxy = ideal_cmpxy;
2019     best_bol = _gvn.transform(new BoolNode(cmpxy, BoolTest::lt));
2020     // and fall through:
2021   case BoolTest::lt:          // x < y
2022   case BoolTest::le:          // x <= y
2023     answer_if_true  = (want_max ? yvalue : xvalue);
2024     answer_if_false = (want_max ? xvalue : yvalue);
2025     break;
2026   case BoolTest::gt:          // x > y
2027   case BoolTest::ge:          // x >= y
2028     answer_if_true  = (want_max ? xvalue : yvalue);
2029     answer_if_false = (want_max ? yvalue : xvalue);
2030     break;
2031   }
2032 
2033   jint hi, lo;
2034   if (want_max) {
2035     // We can sharpen the minimum.
2036     hi = MAX2(txvalue->_hi, tyvalue->_hi);
2037     lo = MAX2(txvalue->_lo, tyvalue->_lo);
2038   } else {
2039     // We can sharpen the maximum.
2040     hi = MIN2(txvalue->_hi, tyvalue->_hi);
2041     lo = MIN2(txvalue->_lo, tyvalue->_lo);
2042   }
2043 
2044   // Use a flow-free graph structure, to avoid creating excess control edges
2045   // which could hinder other optimizations.
2046   // Since Math.min/max is often used with arraycopy, we want
2047   // tightly_coupled_allocation to be able to see beyond min/max expressions.
2048   Node* cmov = CMoveNode::make(NULL, best_bol,
2049                                answer_if_false, answer_if_true,
2050                                TypeInt::make(lo, hi, widen));
2051 
2052   return _gvn.transform(cmov);
2053 
2054   /*
2055   // This is not as desirable as it may seem, since Min and Max
2056   // nodes do not have a full set of optimizations.
2057   // And they would interfere, anyway, with 'if' optimizations
2058   // and with CMoveI canonical forms.
2059   switch (id) {
2060   case vmIntrinsics::_min:
2061     result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
2062   case vmIntrinsics::_max:
2063     result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
2064   default:
2065     ShouldNotReachHere();
2066   }
2067   */
2068 }
2069 
2070 inline int
2071 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2072   const TypePtr* base_type = TypePtr::NULL_PTR;
2073   if (base != NULL)  base_type = _gvn.type(base)->isa_ptr();
2074   if (base_type == NULL) {
2075     // Unknown type.
2076     return Type::AnyPtr;
2077   } else if (base_type == TypePtr::NULL_PTR) {
2078     // Since this is a NULL+long form, we have to switch to a rawptr.
2079     base   = _gvn.transform(new CastX2PNode(offset));
2080     offset = MakeConX(0);
2081     return Type::RawPtr;
2082   } else if (base_type->base() == Type::RawPtr) {
2083     return Type::RawPtr;
2084   } else if (base_type->isa_oopptr()) {
2085     // Base is never null => always a heap address.
2086     if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2087       return Type::OopPtr;
2088     }
2089     // Offset is small => always a heap address.
2090     const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2091     if (offset_type != NULL &&
2092         base_type->offset() == 0 &&     // (should always be?)
2093         offset_type->_lo >= 0 &&
2094         !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2095       return Type::OopPtr;
2096     } else if (type == T_OBJECT) {
2097       // off heap access to an oop doesn't make any sense. Has to be on
2098       // heap.
2099       return Type::OopPtr;
2100     }
2101     // Otherwise, it might either be oop+off or NULL+addr.
2102     return Type::AnyPtr;
2103   } else {
2104     // No information:
2105     return Type::AnyPtr;
2106   }
2107 }
2108 
2109 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2110   Node* uncasted_base = base;
2111   int kind = classify_unsafe_addr(uncasted_base, offset, type);
2112   if (kind == Type::RawPtr) {
2113     return basic_plus_adr(top(), uncasted_base, offset);
2114   } else if (kind == Type::AnyPtr) {
2115     assert(base == uncasted_base, "unexpected base change");
2116     if (can_cast) {
2117       if (!_gvn.type(base)->speculative_maybe_null() &&
2118           !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2119         // According to profiling, this access is always on
2120         // heap. Casting the base to not null and thus avoiding membars
2121         // around the access should allow better optimizations
2122         Node* null_ctl = top();
2123         base = null_check_oop(base, &null_ctl, true, true, true);
2124         assert(null_ctl->is_top(), "no null control here");
2125         return basic_plus_adr(base, offset);
2126       } else if (_gvn.type(base)->speculative_always_null() &&
2127                  !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2128         // According to profiling, this access is always off
2129         // heap.
2130         base = null_assert(base);
2131         Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2132         offset = MakeConX(0);
2133         return basic_plus_adr(top(), raw_base, offset);
2134       }
2135     }
2136     // We don't know if it's an on heap or off heap access. Fall back
2137     // to raw memory access.
2138     Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2139     return basic_plus_adr(top(), raw, offset);
2140   } else {
2141     assert(base == uncasted_base, "unexpected base change");
2142     // We know it's an on heap access so base can't be null
2143     if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2144       base = must_be_not_null(base, true);
2145     }
2146     return basic_plus_adr(base, offset);
2147   }
2148 }
2149 
2150 //--------------------------inline_number_methods-----------------------------
2151 // inline int     Integer.numberOfLeadingZeros(int)
2152 // inline int        Long.numberOfLeadingZeros(long)
2153 //
2154 // inline int     Integer.numberOfTrailingZeros(int)
2155 // inline int        Long.numberOfTrailingZeros(long)
2156 //
2157 // inline int     Integer.bitCount(int)
2158 // inline int        Long.bitCount(long)
2159 //
2160 // inline char  Character.reverseBytes(char)
2161 // inline short     Short.reverseBytes(short)
2162 // inline int     Integer.reverseBytes(int)
2163 // inline long       Long.reverseBytes(long)
2164 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2165   Node* arg = argument(0);
2166   Node* n = NULL;
2167   switch (id) {
2168   case vmIntrinsics::_numberOfLeadingZeros_i:   n = new CountLeadingZerosINode( arg);  break;
2169   case vmIntrinsics::_numberOfLeadingZeros_l:   n = new CountLeadingZerosLNode( arg);  break;
2170   case vmIntrinsics::_numberOfTrailingZeros_i:  n = new CountTrailingZerosINode(arg);  break;
2171   case vmIntrinsics::_numberOfTrailingZeros_l:  n = new CountTrailingZerosLNode(arg);  break;
2172   case vmIntrinsics::_bitCount_i:               n = new PopCountINode(          arg);  break;
2173   case vmIntrinsics::_bitCount_l:               n = new PopCountLNode(          arg);  break;
2174   case vmIntrinsics::_reverseBytes_c:           n = new ReverseBytesUSNode(0,   arg);  break;
2175   case vmIntrinsics::_reverseBytes_s:           n = new ReverseBytesSNode( 0,   arg);  break;
2176   case vmIntrinsics::_reverseBytes_i:           n = new ReverseBytesINode( 0,   arg);  break;
2177   case vmIntrinsics::_reverseBytes_l:           n = new ReverseBytesLNode( 0,   arg);  break;
2178   default:  fatal_unexpected_iid(id);  break;
2179   }
2180   set_result(_gvn.transform(n));
2181   return true;
2182 }
2183 
2184 //----------------------------inline_unsafe_access----------------------------
2185 
2186 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2187   // Attempt to infer a sharper value type from the offset and base type.
2188   ciKlass* sharpened_klass = NULL;
2189 
2190   // See if it is an instance field, with an object type.
2191   if (alias_type->field() != NULL) {
2192     if (alias_type->field()->type()->is_klass()) {
2193       sharpened_klass = alias_type->field()->type()->as_klass();
2194     }
2195   }
2196 
2197   // See if it is a narrow oop array.
2198   if (adr_type->isa_aryptr()) {
2199     if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2200       const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2201       if (elem_type != NULL) {
2202         sharpened_klass = elem_type->klass();
2203       }
2204     }
2205   }
2206 
2207   // The sharpened class might be unloaded if there is no class loader
2208   // contraint in place.
2209   if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2210     const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);
2211 
2212 #ifndef PRODUCT
2213     if (C->print_intrinsics() || C->print_inlining()) {
2214       tty->print("  from base type:  ");  adr_type->dump(); tty->cr();
2215       tty->print("  sharpened value: ");  tjp->dump();      tty->cr();
2216     }
2217 #endif
2218     // Sharpen the value type.
2219     return tjp;
2220   }
2221   return NULL;
2222 }
2223 
2224 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2225   switch (kind) {
2226       case Relaxed:
2227         return MO_UNORDERED;
2228       case Opaque:
2229         return MO_RELAXED;
2230       case Acquire:
2231         return MO_ACQUIRE;
2232       case Release:
2233         return MO_RELEASE;
2234       case Volatile:
2235         return MO_SEQ_CST;
2236       default:
2237         ShouldNotReachHere();
2238         return 0;
2239   }
2240 }
2241 
2242 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2243   if (callee()->is_static())  return false;  // caller must have the capability!
2244   DecoratorSet decorators = C2_UNSAFE_ACCESS;
2245   guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2246   guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2247   assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2248 
2249   if (is_reference_type(type)) {
2250     decorators |= ON_UNKNOWN_OOP_REF;
2251   }
2252 
2253   if (unaligned) {
2254     decorators |= C2_UNALIGNED;
2255   }
2256 
2257 #ifndef PRODUCT
2258   {
2259     ResourceMark rm;
2260     // Check the signatures.
2261     ciSignature* sig = callee()->signature();
2262 #ifdef ASSERT
2263     if (!is_store) {
2264       // Object getReference(Object base, int/long offset), etc.
2265       BasicType rtype = sig->return_type()->basic_type();
2266       assert(rtype == type, "getter must return the expected value");
2267       assert(sig->count() == 2, "oop getter has 2 arguments");
2268       assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2269       assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2270     } else {
2271       // void putReference(Object base, int/long offset, Object x), etc.
2272       assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2273       assert(sig->count() == 3, "oop putter has 3 arguments");
2274       assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2275       assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2276       BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2277       assert(vtype == type, "putter must accept the expected value");
2278     }
2279 #endif // ASSERT
2280  }
2281 #endif //PRODUCT
2282 
2283   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2284 
2285   Node* receiver = argument(0);  // type: oop
2286 
2287   // Build address expression.
2288   Node* heap_base_oop = top();
2289 
2290   // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2291   Node* base = argument(1);  // type: oop
2292   // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2293   Node* offset = argument(2);  // type: long
2294   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2295   // to be plain byte offsets, which are also the same as those accepted
2296   // by oopDesc::field_addr.
2297   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2298          "fieldOffset must be byte-scaled");
2299   // 32-bit machines ignore the high half!
2300   offset = ConvL2X(offset);
2301 
2302   // Save state and restore on bailout
2303   uint old_sp = sp();
2304   SafePointNode* old_map = clone_map();
2305 
2306   Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2307 
2308   if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2309     if (type != T_OBJECT) {
2310       decorators |= IN_NATIVE; // off-heap primitive access
2311     } else {
2312       set_map(old_map);
2313       set_sp(old_sp);
2314       return false; // off-heap oop accesses are not supported
2315     }
2316   } else {
2317     heap_base_oop = base; // on-heap or mixed access
2318   }
2319 
2320   // Can base be NULL? Otherwise, always on-heap access.
2321   bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2322 
2323   if (!can_access_non_heap) {
2324     decorators |= IN_HEAP;
2325   }
2326 
2327   Node* val = is_store ? argument(4) : NULL;
2328 
2329   const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2330   if (adr_type == TypePtr::NULL_PTR) {
2331     set_map(old_map);
2332     set_sp(old_sp);
2333     return false; // off-heap access with zero address
2334   }
2335 
2336   // Try to categorize the address.
2337   Compile::AliasType* alias_type = C->alias_type(adr_type);
2338   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2339 
2340   if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2341       alias_type->adr_type() == TypeAryPtr::RANGE) {
2342     set_map(old_map);
2343     set_sp(old_sp);
2344     return false; // not supported
2345   }
2346 
2347   bool mismatched = false;
2348   BasicType bt = alias_type->basic_type();
2349   if (bt != T_ILLEGAL) {
2350     assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2351     if (bt == T_BYTE && adr_type->isa_aryptr()) {
2352       // Alias type doesn't differentiate between byte[] and boolean[]).
2353       // Use address type to get the element type.
2354       bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2355     }
2356     if (bt == T_ARRAY || bt == T_NARROWOOP) {
2357       // accessing an array field with getReference is not a mismatch
2358       bt = T_OBJECT;
2359     }
2360     if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2361       // Don't intrinsify mismatched object accesses
2362       set_map(old_map);
2363       set_sp(old_sp);
2364       return false;
2365     }
2366     mismatched = (bt != type);
2367   } else if (alias_type->adr_type()->isa_oopptr()) {
2368     mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2369   }
2370 
2371   old_map->destruct(&_gvn);
2372   assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2373 
2374   if (mismatched) {
2375     decorators |= C2_MISMATCHED;
2376   }
2377 
2378   // First guess at the value type.
2379   const Type *value_type = Type::get_const_basic_type(type);
2380 
2381   // Figure out the memory ordering.
2382   decorators |= mo_decorator_for_access_kind(kind);
2383 
2384   if (!is_store && type == T_OBJECT) {
2385     const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2386     if (tjp != NULL) {
2387       value_type = tjp;
2388     }
2389   }
2390 
2391   receiver = null_check(receiver);
2392   if (stopped()) {
2393     return true;
2394   }
2395   // Heap pointers get a null-check from the interpreter,
2396   // as a courtesy.  However, this is not guaranteed by Unsafe,
2397   // and it is not possible to fully distinguish unintended nulls
2398   // from intended ones in this API.
2399 
2400   if (!is_store) {
2401     Node* p = NULL;
2402     // Try to constant fold a load from a constant field
2403     ciField* field = alias_type->field();
2404     if (heap_base_oop != top() && field != NULL && field->is_constant() && !mismatched) {
2405       // final or stable field
2406       p = make_constant_from_field(field, heap_base_oop);
2407     }
2408 
2409     if (p == NULL) { // Could not constant fold the load
2410       p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2411       // Normalize the value returned by getBoolean in the following cases
2412       if (type == T_BOOLEAN &&
2413           (mismatched ||
2414            heap_base_oop == top() ||                  // - heap_base_oop is NULL or
2415            (can_access_non_heap && field == NULL))    // - heap_base_oop is potentially NULL
2416                                                       //   and the unsafe access is made to large offset
2417                                                       //   (i.e., larger than the maximum offset necessary for any
2418                                                       //   field access)
2419             ) {
2420           IdealKit ideal = IdealKit(this);
2421 #define __ ideal.
2422           IdealVariable normalized_result(ideal);
2423           __ declarations_done();
2424           __ set(normalized_result, p);
2425           __ if_then(p, BoolTest::ne, ideal.ConI(0));
2426           __ set(normalized_result, ideal.ConI(1));
2427           ideal.end_if();
2428           final_sync(ideal);
2429           p = __ value(normalized_result);
2430 #undef __
2431       }
2432     }
2433     if (type == T_ADDRESS) {
2434       p = gvn().transform(new CastP2XNode(NULL, p));
2435       p = ConvX2UL(p);
2436     }
2437     // The load node has the control of the preceding MemBarCPUOrder.  All
2438     // following nodes will have the control of the MemBarCPUOrder inserted at
2439     // the end of this method.  So, pushing the load onto the stack at a later
2440     // point is fine.
2441     set_result(p);
2442   } else {
2443     if (bt == T_ADDRESS) {
2444       // Repackage the long as a pointer.
2445       val = ConvL2X(val);
2446       val = gvn().transform(new CastX2PNode(val));
2447     }
2448     access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2449   }
2450 
2451   return true;
2452 }
2453 
2454 //----------------------------inline_unsafe_load_store----------------------------
2455 // This method serves a couple of different customers (depending on LoadStoreKind):
2456 //
2457 // LS_cmp_swap:
2458 //
2459 //   boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2460 //   boolean compareAndSetInt(   Object o, long offset, int    expected, int    x);
2461 //   boolean compareAndSetLong(  Object o, long offset, long   expected, long   x);
2462 //
2463 // LS_cmp_swap_weak:
2464 //
2465 //   boolean weakCompareAndSetReference(       Object o, long offset, Object expected, Object x);
2466 //   boolean weakCompareAndSetReferencePlain(  Object o, long offset, Object expected, Object x);
2467 //   boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2468 //   boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2469 //
2470 //   boolean weakCompareAndSetInt(          Object o, long offset, int    expected, int    x);
2471 //   boolean weakCompareAndSetIntPlain(     Object o, long offset, int    expected, int    x);
2472 //   boolean weakCompareAndSetIntAcquire(   Object o, long offset, int    expected, int    x);
2473 //   boolean weakCompareAndSetIntRelease(   Object o, long offset, int    expected, int    x);
2474 //
2475 //   boolean weakCompareAndSetLong(         Object o, long offset, long   expected, long   x);
2476 //   boolean weakCompareAndSetLongPlain(    Object o, long offset, long   expected, long   x);
2477 //   boolean weakCompareAndSetLongAcquire(  Object o, long offset, long   expected, long   x);
2478 //   boolean weakCompareAndSetLongRelease(  Object o, long offset, long   expected, long   x);
2479 //
2480 // LS_cmp_exchange:
2481 //
2482 //   Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2483 //   Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2484 //   Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2485 //
2486 //   Object compareAndExchangeIntVolatile(   Object o, long offset, Object expected, Object x);
2487 //   Object compareAndExchangeIntAcquire(    Object o, long offset, Object expected, Object x);
2488 //   Object compareAndExchangeIntRelease(    Object o, long offset, Object expected, Object x);
2489 //
2490 //   Object compareAndExchangeLongVolatile(  Object o, long offset, Object expected, Object x);
2491 //   Object compareAndExchangeLongAcquire(   Object o, long offset, Object expected, Object x);
2492 //   Object compareAndExchangeLongRelease(   Object o, long offset, Object expected, Object x);
2493 //
2494 // LS_get_add:
2495 //
2496 //   int  getAndAddInt( Object o, long offset, int  delta)
2497 //   long getAndAddLong(Object o, long offset, long delta)
2498 //
2499 // LS_get_set:
2500 //
2501 //   int    getAndSet(Object o, long offset, int    newValue)
2502 //   long   getAndSet(Object o, long offset, long   newValue)
2503 //   Object getAndSet(Object o, long offset, Object newValue)
2504 //
2505 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2506   // This basic scheme here is the same as inline_unsafe_access, but
2507   // differs in enough details that combining them would make the code
2508   // overly confusing.  (This is a true fact! I originally combined
2509   // them, but even I was confused by it!) As much code/comments as
2510   // possible are retained from inline_unsafe_access though to make
2511   // the correspondences clearer. - dl
2512 
2513   if (callee()->is_static())  return false;  // caller must have the capability!
2514 
2515   DecoratorSet decorators = C2_UNSAFE_ACCESS;
2516   decorators |= mo_decorator_for_access_kind(access_kind);
2517 
2518 #ifndef PRODUCT
2519   BasicType rtype;
2520   {
2521     ResourceMark rm;
2522     // Check the signatures.
2523     ciSignature* sig = callee()->signature();
2524     rtype = sig->return_type()->basic_type();
2525     switch(kind) {
2526       case LS_get_add:
2527       case LS_get_set: {
2528       // Check the signatures.
2529 #ifdef ASSERT
2530       assert(rtype == type, "get and set must return the expected type");
2531       assert(sig->count() == 3, "get and set has 3 arguments");
2532       assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2533       assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2534       assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2535       assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2536 #endif // ASSERT
2537         break;
2538       }
2539       case LS_cmp_swap:
2540       case LS_cmp_swap_weak: {
2541       // Check the signatures.
2542 #ifdef ASSERT
2543       assert(rtype == T_BOOLEAN, "CAS must return boolean");
2544       assert(sig->count() == 4, "CAS has 4 arguments");
2545       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2546       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2547 #endif // ASSERT
2548         break;
2549       }
2550       case LS_cmp_exchange: {
2551       // Check the signatures.
2552 #ifdef ASSERT
2553       assert(rtype == type, "CAS must return the expected type");
2554       assert(sig->count() == 4, "CAS has 4 arguments");
2555       assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2556       assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2557 #endif // ASSERT
2558         break;
2559       }
2560       default:
2561         ShouldNotReachHere();
2562     }
2563   }
2564 #endif //PRODUCT
2565 
2566   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
2567 
2568   // Get arguments:
2569   Node* receiver = NULL;
2570   Node* base     = NULL;
2571   Node* offset   = NULL;
2572   Node* oldval   = NULL;
2573   Node* newval   = NULL;
2574   switch(kind) {
2575     case LS_cmp_swap:
2576     case LS_cmp_swap_weak:
2577     case LS_cmp_exchange: {
2578       const bool two_slot_type = type2size[type] == 2;
2579       receiver = argument(0);  // type: oop
2580       base     = argument(1);  // type: oop
2581       offset   = argument(2);  // type: long
2582       oldval   = argument(4);  // type: oop, int, or long
2583       newval   = argument(two_slot_type ? 6 : 5);  // type: oop, int, or long
2584       break;
2585     }
2586     case LS_get_add:
2587     case LS_get_set: {
2588       receiver = argument(0);  // type: oop
2589       base     = argument(1);  // type: oop
2590       offset   = argument(2);  // type: long
2591       oldval   = NULL;
2592       newval   = argument(4);  // type: oop, int, or long
2593       break;
2594     }
2595     default:
2596       ShouldNotReachHere();
2597   }
2598 
2599   // Build field offset expression.
2600   // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2601   // to be plain byte offsets, which are also the same as those accepted
2602   // by oopDesc::field_addr.
2603   assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2604   // 32-bit machines ignore the high half of long offsets
2605   offset = ConvL2X(offset);
2606   // Save state and restore on bailout
2607   uint old_sp = sp();
2608   SafePointNode* old_map = clone_map();
2609   Node* adr = make_unsafe_address(base, offset,type, false);
2610   const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2611 
2612   Compile::AliasType* alias_type = C->alias_type(adr_type);
2613   BasicType bt = alias_type->basic_type();
2614   if (bt != T_ILLEGAL &&
2615       (is_reference_type(bt) != (type == T_OBJECT))) {
2616     // Don't intrinsify mismatched object accesses.
2617     set_map(old_map);
2618     set_sp(old_sp);
2619     return false;
2620   }
2621 
2622   old_map->destruct(&_gvn);
2623 
2624   // For CAS, unlike inline_unsafe_access, there seems no point in
2625   // trying to refine types. Just use the coarse types here.
2626   assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2627   const Type *value_type = Type::get_const_basic_type(type);
2628 
2629   switch (kind) {
2630     case LS_get_set:
2631     case LS_cmp_exchange: {
2632       if (type == T_OBJECT) {
2633         const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2634         if (tjp != NULL) {
2635           value_type = tjp;
2636         }
2637       }
2638       break;
2639     }
2640     case LS_cmp_swap:
2641     case LS_cmp_swap_weak:
2642     case LS_get_add:
2643       break;
2644     default:
2645       ShouldNotReachHere();
2646   }
2647 
2648   // Null check receiver.
2649   receiver = null_check(receiver);
2650   if (stopped()) {
2651     return true;
2652   }
2653 
2654   int alias_idx = C->get_alias_index(adr_type);
2655 
2656   if (is_reference_type(type)) {
2657     decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2658 
2659     // Transformation of a value which could be NULL pointer (CastPP #NULL)
2660     // could be delayed during Parse (for example, in adjust_map_after_if()).
2661     // Execute transformation here to avoid barrier generation in such case.
2662     if (_gvn.type(newval) == TypePtr::NULL_PTR)
2663       newval = _gvn.makecon(TypePtr::NULL_PTR);
2664 
2665     if (oldval != NULL && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2666       // Refine the value to a null constant, when it is known to be null
2667       oldval = _gvn.makecon(TypePtr::NULL_PTR);
2668     }
2669   }
2670 
2671   Node* result = NULL;
2672   switch (kind) {
2673     case LS_cmp_exchange: {
2674       result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2675                                             oldval, newval, value_type, type, decorators);
2676       break;
2677     }
2678     case LS_cmp_swap_weak:
2679       decorators |= C2_WEAK_CMPXCHG;
2680     case LS_cmp_swap: {
2681       result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2682                                              oldval, newval, value_type, type, decorators);
2683       break;
2684     }
2685     case LS_get_set: {
2686       result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2687                                      newval, value_type, type, decorators);
2688       break;
2689     }
2690     case LS_get_add: {
2691       result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2692                                     newval, value_type, type, decorators);
2693       break;
2694     }
2695     default:
2696       ShouldNotReachHere();
2697   }
2698 
2699   assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2700   set_result(result);
2701   return true;
2702 }
2703 
2704 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2705   // Regardless of form, don't allow previous ld/st to move down,
2706   // then issue acquire, release, or volatile mem_bar.
2707   insert_mem_bar(Op_MemBarCPUOrder);
2708   switch(id) {
2709     case vmIntrinsics::_loadFence:
2710       insert_mem_bar(Op_LoadFence);
2711       return true;
2712     case vmIntrinsics::_storeFence:
2713       insert_mem_bar(Op_StoreFence);
2714       return true;
2715     case vmIntrinsics::_storeStoreFence:
2716       insert_mem_bar(Op_StoreStoreFence);
2717       return true;
2718     case vmIntrinsics::_fullFence:
2719       insert_mem_bar(Op_MemBarVolatile);
2720       return true;
2721     default:
2722       fatal_unexpected_iid(id);
2723       return false;
2724   }
2725 }
2726 
2727 bool LibraryCallKit::inline_onspinwait() {
2728   insert_mem_bar(Op_OnSpinWait);
2729   return true;
2730 }
2731 
2732 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2733   if (!kls->is_Con()) {
2734     return true;
2735   }
2736   const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
2737   if (klsptr == NULL) {
2738     return true;
2739   }
2740   ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
2741   // don't need a guard for a klass that is already initialized
2742   return !ik->is_initialized();
2743 }
2744 
2745 //----------------------------inline_unsafe_writeback0-------------------------
2746 // public native void Unsafe.writeback0(long address)
2747 bool LibraryCallKit::inline_unsafe_writeback0() {
2748   if (!Matcher::has_match_rule(Op_CacheWB)) {
2749     return false;
2750   }
2751 #ifndef PRODUCT
2752   assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2753   assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2754   ciSignature* sig = callee()->signature();
2755   assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2756 #endif
2757   null_check_receiver();  // null-check, then ignore
2758   Node *addr = argument(1);
2759   addr = new CastX2PNode(addr);
2760   addr = _gvn.transform(addr);
2761   Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2762   flush = _gvn.transform(flush);
2763   set_memory(flush, TypeRawPtr::BOTTOM);
2764   return true;
2765 }
2766 
2767 //----------------------------inline_unsafe_writeback0-------------------------
2768 // public native void Unsafe.writeback0(long address)
2769 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2770   if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2771     return false;
2772   }
2773   if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2774     return false;
2775   }
2776 #ifndef PRODUCT
2777   assert(Matcher::has_match_rule(Op_CacheWB),
2778          (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2779                 : "found match rule for CacheWBPostSync but not CacheWB"));
2780 
2781 #endif
2782   null_check_receiver();  // null-check, then ignore
2783   Node *sync;
2784   if (is_pre) {
2785     sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2786   } else {
2787     sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2788   }
2789   sync = _gvn.transform(sync);
2790   set_memory(sync, TypeRawPtr::BOTTOM);
2791   return true;
2792 }
2793 
2794 //----------------------------inline_unsafe_allocate---------------------------
2795 // public native Object Unsafe.allocateInstance(Class<?> cls);
2796 bool LibraryCallKit::inline_unsafe_allocate() {
2797   if (callee()->is_static())  return false;  // caller must have the capability!
2798 
2799   null_check_receiver();  // null-check, then ignore
2800   Node* cls = null_check(argument(1));
2801   if (stopped())  return true;
2802 
2803   Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
2804   kls = null_check(kls);
2805   if (stopped())  return true;  // argument was like int.class
2806 
2807   Node* test = NULL;
2808   if (LibraryCallKit::klass_needs_init_guard(kls)) {
2809     // Note:  The argument might still be an illegal value like
2810     // Serializable.class or Object[].class.   The runtime will handle it.
2811     // But we must make an explicit check for initialization.
2812     Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2813     // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2814     // can generate code to load it as unsigned byte.
2815     Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2816     Node* bits = intcon(InstanceKlass::fully_initialized);
2817     test = _gvn.transform(new SubINode(inst, bits));
2818     // The 'test' is non-zero if we need to take a slow path.
2819   }
2820 
2821   Node* obj = new_instance(kls, test);
2822   set_result(obj);
2823   return true;
2824 }
2825 
2826 //------------------------inline_native_time_funcs--------------
2827 // inline code for System.currentTimeMillis() and System.nanoTime()
2828 // these have the same type and signature
2829 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2830   const TypeFunc* tf = OptoRuntime::void_long_Type();
2831   const TypePtr* no_memory_effects = NULL;
2832   Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2833   Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2834 #ifdef ASSERT
2835   Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2836   assert(value_top == top(), "second value must be top");
2837 #endif
2838   set_result(value);
2839   return true;
2840 }
2841 
2842 #ifdef JFR_HAVE_INTRINSICS
2843 
2844 /**
2845  * if oop->klass != null
2846  *   // normal class
2847  *   epoch = _epoch_state ? 2 : 1
2848  *   if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2849  *     ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2850  *   }
2851  *   id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2852  * else
2853  *   // primitive class
2854  *   if oop->array_klass != null
2855  *     id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2856  *   else
2857  *     id = LAST_TYPE_ID + 1 // void class path
2858  *   if (!signaled)
2859  *     signaled = true
2860  */
2861 bool LibraryCallKit::inline_native_classID() {
2862   Node* cls = argument(0);
2863 
2864   IdealKit ideal(this);
2865 #define __ ideal.
2866   IdealVariable result(ideal); __ declarations_done();
2867   Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2868                                                  basic_plus_adr(cls, java_lang_Class::klass_offset()),
2869                                                  TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2870 
2871 
2872   __ if_then(kls, BoolTest::ne, null()); {
2873     Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2874     Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2875 
2876     Node* epoch_address = makecon(TypeRawPtr::make(Jfr::epoch_address()));
2877     Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2878     epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
2879     Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
2880     mask = _gvn.transform(new OrLNode(mask, epoch));
2881     Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
2882 
2883     float unlikely  = PROB_UNLIKELY(0.999);
2884     __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
2885       sync_kit(ideal);
2886       make_runtime_call(RC_LEAF,
2887                         OptoRuntime::get_class_id_intrinsic_Type(),
2888                         CAST_FROM_FN_PTR(address, Jfr::get_class_id_intrinsic),
2889                         "get_class_id_intrinsic",
2890                         TypePtr::BOTTOM,
2891                         kls);
2892       ideal.sync_kit(this);
2893     } __ end_if();
2894 
2895     ideal.set(result,  _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
2896   } __ else_(); {
2897     Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(),
2898                                                    basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
2899                                                    TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
2900     __ if_then(array_kls, BoolTest::ne, null()); {
2901       Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
2902       Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
2903       Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
2904       ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
2905     } __ else_(); {
2906       // void class case
2907       ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
2908     } __ end_if();
2909 
2910     Node* signaled_flag_address = makecon(TypeRawPtr::make(Jfr::signal_address()));
2911     Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
2912     __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
2913       ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
2914     } __ end_if();
2915   } __ end_if();
2916 
2917   final_sync(ideal);
2918   set_result(ideal.value(result));
2919 #undef __
2920   return true;
2921 }
2922 
2923 bool LibraryCallKit::inline_native_getEventWriter() {
2924   Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
2925 
2926   Node* jobj_ptr = basic_plus_adr(top(), tls_ptr,
2927                                   in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
2928 
2929   Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
2930 
2931   Node* jobj_cmp_null = _gvn.transform( new CmpPNode(jobj, null()) );
2932   Node* test_jobj_eq_null  = _gvn.transform( new BoolNode(jobj_cmp_null, BoolTest::eq) );
2933 
2934   IfNode* iff_jobj_null =
2935     create_and_map_if(control(), test_jobj_eq_null, PROB_MIN, COUNT_UNKNOWN);
2936 
2937   enum { _normal_path = 1,
2938          _null_path = 2,
2939          PATH_LIMIT };
2940 
2941   RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
2942   PhiNode*    result_val = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
2943 
2944   Node* jobj_is_null = _gvn.transform(new IfTrueNode(iff_jobj_null));
2945   result_rgn->init_req(_null_path, jobj_is_null);
2946   result_val->init_req(_null_path, null());
2947 
2948   Node* jobj_is_not_null = _gvn.transform(new IfFalseNode(iff_jobj_null));
2949   set_control(jobj_is_not_null);
2950   Node* res = access_load(jobj, TypeInstPtr::NOTNULL, T_OBJECT,
2951                           IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
2952   result_rgn->init_req(_normal_path, control());
2953   result_val->init_req(_normal_path, res);
2954 
2955   set_result(result_rgn, result_val);
2956 
2957   return true;
2958 }
2959 
2960 #endif // JFR_HAVE_INTRINSICS
2961 
2962 //------------------------inline_native_currentThread------------------
2963 bool LibraryCallKit::inline_native_currentThread() {
2964   Node* junk = NULL;
2965   set_result(generate_current_thread(junk));
2966   return true;
2967 }
2968 
2969 //---------------------------load_mirror_from_klass----------------------------
2970 // Given a klass oop, load its java mirror (a java.lang.Class oop).
2971 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
2972   Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
2973   Node* load = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
2974   // mirror = ((OopHandle)mirror)->resolve();
2975   return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
2976 }
2977 
2978 //-----------------------load_klass_from_mirror_common-------------------------
2979 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
2980 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
2981 // and branch to the given path on the region.
2982 // If never_see_null, take an uncommon trap on null, so we can optimistically
2983 // compile for the non-null case.
2984 // If the region is NULL, force never_see_null = true.
2985 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
2986                                                     bool never_see_null,
2987                                                     RegionNode* region,
2988                                                     int null_path,
2989                                                     int offset) {
2990   if (region == NULL)  never_see_null = true;
2991   Node* p = basic_plus_adr(mirror, offset);
2992   const TypeKlassPtr*  kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
2993   Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
2994   Node* null_ctl = top();
2995   kls = null_check_oop(kls, &null_ctl, never_see_null);
2996   if (region != NULL) {
2997     // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
2998     region->init_req(null_path, null_ctl);
2999   } else {
3000     assert(null_ctl == top(), "no loose ends");
3001   }
3002   return kls;
3003 }
3004 
3005 //--------------------(inline_native_Class_query helpers)---------------------
3006 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3007 // Fall through if (mods & mask) == bits, take the guard otherwise.
3008 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3009   // Branch around if the given klass has the given modifier bit set.
3010   // Like generate_guard, adds a new path onto the region.
3011   Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3012   Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT, MemNode::unordered);
3013   Node* mask = intcon(modifier_mask);
3014   Node* bits = intcon(modifier_bits);
3015   Node* mbit = _gvn.transform(new AndINode(mods, mask));
3016   Node* cmp  = _gvn.transform(new CmpINode(mbit, bits));
3017   Node* bol  = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3018   return generate_fair_guard(bol, region);
3019 }
3020 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3021   return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3022 }
3023 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3024   return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3025 }
3026 
3027 //-------------------------inline_native_Class_query-------------------
3028 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3029   const Type* return_type = TypeInt::BOOL;
3030   Node* prim_return_value = top();  // what happens if it's a primitive class?
3031   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3032   bool expect_prim = false;     // most of these guys expect to work on refs
3033 
3034   enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3035 
3036   Node* mirror = argument(0);
3037   Node* obj    = top();
3038 
3039   switch (id) {
3040   case vmIntrinsics::_isInstance:
3041     // nothing is an instance of a primitive type
3042     prim_return_value = intcon(0);
3043     obj = argument(1);
3044     break;
3045   case vmIntrinsics::_getModifiers:
3046     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3047     assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3048     return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3049     break;
3050   case vmIntrinsics::_isInterface:
3051     prim_return_value = intcon(0);
3052     break;
3053   case vmIntrinsics::_isArray:
3054     prim_return_value = intcon(0);
3055     expect_prim = true;  // cf. ObjectStreamClass.getClassSignature
3056     break;
3057   case vmIntrinsics::_isPrimitive:
3058     prim_return_value = intcon(1);
3059     expect_prim = true;  // obviously
3060     break;
3061   case vmIntrinsics::_isHidden:
3062     prim_return_value = intcon(0);
3063     break;
3064   case vmIntrinsics::_getSuperclass:
3065     prim_return_value = null();
3066     return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3067     break;
3068   case vmIntrinsics::_getClassAccessFlags:
3069     prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3070     return_type = TypeInt::INT;  // not bool!  6297094
3071     break;
3072   default:
3073     fatal_unexpected_iid(id);
3074     break;
3075   }
3076 
3077   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3078   if (mirror_con == NULL)  return false;  // cannot happen?
3079 
3080 #ifndef PRODUCT
3081   if (C->print_intrinsics() || C->print_inlining()) {
3082     ciType* k = mirror_con->java_mirror_type();
3083     if (k) {
3084       tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3085       k->print_name();
3086       tty->cr();
3087     }
3088   }
3089 #endif
3090 
3091   // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3092   RegionNode* region = new RegionNode(PATH_LIMIT);
3093   record_for_igvn(region);
3094   PhiNode* phi = new PhiNode(region, return_type);
3095 
3096   // The mirror will never be null of Reflection.getClassAccessFlags, however
3097   // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3098   // if it is. See bug 4774291.
3099 
3100   // For Reflection.getClassAccessFlags(), the null check occurs in
3101   // the wrong place; see inline_unsafe_access(), above, for a similar
3102   // situation.
3103   mirror = null_check(mirror);
3104   // If mirror or obj is dead, only null-path is taken.
3105   if (stopped())  return true;
3106 
3107   if (expect_prim)  never_see_null = false;  // expect nulls (meaning prims)
3108 
3109   // Now load the mirror's klass metaobject, and null-check it.
3110   // Side-effects region with the control path if the klass is null.
3111   Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3112   // If kls is null, we have a primitive mirror.
3113   phi->init_req(_prim_path, prim_return_value);
3114   if (stopped()) { set_result(region, phi); return true; }
3115   bool safe_for_replace = (region->in(_prim_path) == top());
3116 
3117   Node* p;  // handy temp
3118   Node* null_ctl;
3119 
3120   // Now that we have the non-null klass, we can perform the real query.
3121   // For constant classes, the query will constant-fold in LoadNode::Value.
3122   Node* query_value = top();
3123   switch (id) {
3124   case vmIntrinsics::_isInstance:
3125     // nothing is an instance of a primitive type
3126     query_value = gen_instanceof(obj, kls, safe_for_replace);
3127     break;
3128 
3129   case vmIntrinsics::_getModifiers:
3130     p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3131     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3132     break;
3133 
3134   case vmIntrinsics::_isInterface:
3135     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3136     if (generate_interface_guard(kls, region) != NULL)
3137       // A guard was added.  If the guard is taken, it was an interface.
3138       phi->add_req(intcon(1));
3139     // If we fall through, it's a plain class.
3140     query_value = intcon(0);
3141     break;
3142 
3143   case vmIntrinsics::_isArray:
3144     // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3145     if (generate_array_guard(kls, region) != NULL)
3146       // A guard was added.  If the guard is taken, it was an array.
3147       phi->add_req(intcon(1));
3148     // If we fall through, it's a plain class.
3149     query_value = intcon(0);
3150     break;
3151 
3152   case vmIntrinsics::_isPrimitive:
3153     query_value = intcon(0); // "normal" path produces false
3154     break;
3155 
3156   case vmIntrinsics::_isHidden:
3157     // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3158     if (generate_hidden_class_guard(kls, region) != NULL)
3159       // A guard was added.  If the guard is taken, it was an hidden class.
3160       phi->add_req(intcon(1));
3161     // If we fall through, it's a plain class.
3162     query_value = intcon(0);
3163     break;
3164 
3165 
3166   case vmIntrinsics::_getSuperclass:
3167     // The rules here are somewhat unfortunate, but we can still do better
3168     // with random logic than with a JNI call.
3169     // Interfaces store null or Object as _super, but must report null.
3170     // Arrays store an intermediate super as _super, but must report Object.
3171     // Other types can report the actual _super.
3172     // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3173     if (generate_interface_guard(kls, region) != NULL)
3174       // A guard was added.  If the guard is taken, it was an interface.
3175       phi->add_req(null());
3176     if (generate_array_guard(kls, region) != NULL)
3177       // A guard was added.  If the guard is taken, it was an array.
3178       phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3179     // If we fall through, it's a plain class.  Get its _super.
3180     p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3181     kls = _gvn.transform(LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3182     null_ctl = top();
3183     kls = null_check_oop(kls, &null_ctl);
3184     if (null_ctl != top()) {
3185       // If the guard is taken, Object.superClass is null (both klass and mirror).
3186       region->add_req(null_ctl);
3187       phi   ->add_req(null());
3188     }
3189     if (!stopped()) {
3190       query_value = load_mirror_from_klass(kls);
3191     }
3192     break;
3193 
3194   case vmIntrinsics::_getClassAccessFlags:
3195     p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3196     query_value = make_load(NULL, p, TypeInt::INT, T_INT, MemNode::unordered);
3197     break;
3198 
3199   default:
3200     fatal_unexpected_iid(id);
3201     break;
3202   }
3203 
3204   // Fall-through is the normal case of a query to a real class.
3205   phi->init_req(1, query_value);
3206   region->init_req(1, control());
3207 
3208   C->set_has_split_ifs(true); // Has chance for split-if optimization
3209   set_result(region, phi);
3210   return true;
3211 }
3212 
3213 //-------------------------inline_Class_cast-------------------
3214 bool LibraryCallKit::inline_Class_cast() {
3215   Node* mirror = argument(0); // Class
3216   Node* obj    = argument(1);
3217   const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3218   if (mirror_con == NULL) {
3219     return false;  // dead path (mirror->is_top()).
3220   }
3221   if (obj == NULL || obj->is_top()) {
3222     return false;  // dead path
3223   }
3224   const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3225 
3226   // First, see if Class.cast() can be folded statically.
3227   // java_mirror_type() returns non-null for compile-time Class constants.
3228   ciType* tm = mirror_con->java_mirror_type();
3229   if (tm != NULL && tm->is_klass() &&
3230       tp != NULL && tp->klass() != NULL) {
3231     if (!tp->klass()->is_loaded()) {
3232       // Don't use intrinsic when class is not loaded.
3233       return false;
3234     } else {
3235       int static_res = C->static_subtype_check(tm->as_klass(), tp->klass());
3236       if (static_res == Compile::SSC_always_true) {
3237         // isInstance() is true - fold the code.
3238         set_result(obj);
3239         return true;
3240       } else if (static_res == Compile::SSC_always_false) {
3241         // Don't use intrinsic, have to throw ClassCastException.
3242         // If the reference is null, the non-intrinsic bytecode will
3243         // be optimized appropriately.
3244         return false;
3245       }
3246     }
3247   }
3248 
3249   // Bailout intrinsic and do normal inlining if exception path is frequent.
3250   if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3251     return false;
3252   }
3253 
3254   // Generate dynamic checks.
3255   // Class.cast() is java implementation of _checkcast bytecode.
3256   // Do checkcast (Parse::do_checkcast()) optimizations here.
3257 
3258   mirror = null_check(mirror);
3259   // If mirror is dead, only null-path is taken.
3260   if (stopped()) {
3261     return true;
3262   }
3263 
3264   // Not-subtype or the mirror's klass ptr is NULL (in case it is a primitive).
3265   enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3266   RegionNode* region = new RegionNode(PATH_LIMIT);
3267   record_for_igvn(region);
3268 
3269   // Now load the mirror's klass metaobject, and null-check it.
3270   // If kls is null, we have a primitive mirror and
3271   // nothing is an instance of a primitive type.
3272   Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3273 
3274   Node* res = top();
3275   if (!stopped()) {
3276     Node* bad_type_ctrl = top();
3277     // Do checkcast optimizations.
3278     res = gen_checkcast(obj, kls, &bad_type_ctrl);
3279     region->init_req(_bad_type_path, bad_type_ctrl);
3280   }
3281   if (region->in(_prim_path) != top() ||
3282       region->in(_bad_type_path) != top()) {
3283     // Let Interpreter throw ClassCastException.
3284     PreserveJVMState pjvms(this);
3285     set_control(_gvn.transform(region));
3286     uncommon_trap(Deoptimization::Reason_intrinsic,
3287                   Deoptimization::Action_maybe_recompile);
3288   }
3289   if (!stopped()) {
3290     set_result(res);
3291   }
3292   return true;
3293 }
3294 
3295 
3296 //--------------------------inline_native_subtype_check------------------------
3297 // This intrinsic takes the JNI calls out of the heart of
3298 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
3299 bool LibraryCallKit::inline_native_subtype_check() {
3300   // Pull both arguments off the stack.
3301   Node* args[2];                // two java.lang.Class mirrors: superc, subc
3302   args[0] = argument(0);
3303   args[1] = argument(1);
3304   Node* klasses[2];             // corresponding Klasses: superk, subk
3305   klasses[0] = klasses[1] = top();
3306 
3307   enum {
3308     // A full decision tree on {superc is prim, subc is prim}:
3309     _prim_0_path = 1,           // {P,N} => false
3310                                 // {P,P} & superc!=subc => false
3311     _prim_same_path,            // {P,P} & superc==subc => true
3312     _prim_1_path,               // {N,P} => false
3313     _ref_subtype_path,          // {N,N} & subtype check wins => true
3314     _both_ref_path,             // {N,N} & subtype check loses => false
3315     PATH_LIMIT
3316   };
3317 
3318   RegionNode* region = new RegionNode(PATH_LIMIT);
3319   Node*       phi    = new PhiNode(region, TypeInt::BOOL);
3320   record_for_igvn(region);
3321 
3322   const TypePtr* adr_type = TypeRawPtr::BOTTOM;   // memory type of loads
3323   const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3324   int class_klass_offset = java_lang_Class::klass_offset();
3325 
3326   // First null-check both mirrors and load each mirror's klass metaobject.
3327   int which_arg;
3328   for (which_arg = 0; which_arg <= 1; which_arg++) {
3329     Node* arg = args[which_arg];
3330     arg = null_check(arg);
3331     if (stopped())  break;
3332     args[which_arg] = arg;
3333 
3334     Node* p = basic_plus_adr(arg, class_klass_offset);
3335     Node* kls = LoadKlassNode::make(_gvn, NULL, immutable_memory(), p, adr_type, kls_type);
3336     klasses[which_arg] = _gvn.transform(kls);
3337   }
3338 
3339   // Having loaded both klasses, test each for null.
3340   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3341   for (which_arg = 0; which_arg <= 1; which_arg++) {
3342     Node* kls = klasses[which_arg];
3343     Node* null_ctl = top();
3344     kls = null_check_oop(kls, &null_ctl, never_see_null);
3345     int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
3346     region->init_req(prim_path, null_ctl);
3347     if (stopped())  break;
3348     klasses[which_arg] = kls;
3349   }
3350 
3351   if (!stopped()) {
3352     // now we have two reference types, in klasses[0..1]
3353     Node* subk   = klasses[1];  // the argument to isAssignableFrom
3354     Node* superk = klasses[0];  // the receiver
3355     region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
3356     // now we have a successful reference subtype check
3357     region->set_req(_ref_subtype_path, control());
3358   }
3359 
3360   // If both operands are primitive (both klasses null), then
3361   // we must return true when they are identical primitives.
3362   // It is convenient to test this after the first null klass check.
3363   set_control(region->in(_prim_0_path)); // go back to first null check
3364   if (!stopped()) {
3365     // Since superc is primitive, make a guard for the superc==subc case.
3366     Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
3367     Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
3368     generate_guard(bol_eq, region, PROB_FAIR);
3369     if (region->req() == PATH_LIMIT+1) {
3370       // A guard was added.  If the added guard is taken, superc==subc.
3371       region->swap_edges(PATH_LIMIT, _prim_same_path);
3372       region->del_req(PATH_LIMIT);
3373     }
3374     region->set_req(_prim_0_path, control()); // Not equal after all.
3375   }
3376 
3377   // these are the only paths that produce 'true':
3378   phi->set_req(_prim_same_path,   intcon(1));
3379   phi->set_req(_ref_subtype_path, intcon(1));
3380 
3381   // pull together the cases:
3382   assert(region->req() == PATH_LIMIT, "sane region");
3383   for (uint i = 1; i < region->req(); i++) {
3384     Node* ctl = region->in(i);
3385     if (ctl == NULL || ctl == top()) {
3386       region->set_req(i, top());
3387       phi   ->set_req(i, top());
3388     } else if (phi->in(i) == NULL) {
3389       phi->set_req(i, intcon(0)); // all other paths produce 'false'
3390     }
3391   }
3392 
3393   set_control(_gvn.transform(region));
3394   set_result(_gvn.transform(phi));
3395   return true;
3396 }
3397 
3398 //---------------------generate_array_guard_common------------------------
3399 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
3400                                                   bool obj_array, bool not_array) {
3401 
3402   if (stopped()) {
3403     return NULL;
3404   }
3405 
3406   // If obj_array/non_array==false/false:
3407   // Branch around if the given klass is in fact an array (either obj or prim).
3408   // If obj_array/non_array==false/true:
3409   // Branch around if the given klass is not an array klass of any kind.
3410   // If obj_array/non_array==true/true:
3411   // Branch around if the kls is not an oop array (kls is int[], String, etc.)
3412   // If obj_array/non_array==true/false:
3413   // Branch around if the kls is an oop array (Object[] or subtype)
3414   //
3415   // Like generate_guard, adds a new path onto the region.
3416   jint  layout_con = 0;
3417   Node* layout_val = get_layout_helper(kls, layout_con);
3418   if (layout_val == NULL) {
3419     bool query = (obj_array
3420                   ? Klass::layout_helper_is_objArray(layout_con)
3421                   : Klass::layout_helper_is_array(layout_con));
3422     if (query == not_array) {
3423       return NULL;                       // never a branch
3424     } else {                             // always a branch
3425       Node* always_branch = control();
3426       if (region != NULL)
3427         region->add_req(always_branch);
3428       set_control(top());
3429       return always_branch;
3430     }
3431   }
3432   // Now test the correct condition.
3433   jint  nval = (obj_array
3434                 ? (jint)(Klass::_lh_array_tag_type_value
3435                    <<    Klass::_lh_array_tag_shift)
3436                 : Klass::_lh_neutral_value);
3437   Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
3438   BoolTest::mask btest = BoolTest::lt;  // correct for testing is_[obj]array
3439   // invert the test if we are looking for a non-array
3440   if (not_array)  btest = BoolTest(btest).negate();
3441   Node* bol = _gvn.transform(new BoolNode(cmp, btest));
3442   return generate_fair_guard(bol, region);
3443 }
3444 
3445 
3446 //-----------------------inline_native_newArray--------------------------
3447 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
3448 // private        native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
3449 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
3450   Node* mirror;
3451   Node* count_val;
3452   if (uninitialized) {
3453     mirror    = argument(1);
3454     count_val = argument(2);
3455   } else {
3456     mirror    = argument(0);
3457     count_val = argument(1);
3458   }
3459 
3460   mirror = null_check(mirror);
3461   // If mirror or obj is dead, only null-path is taken.
3462   if (stopped())  return true;
3463 
3464   enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3465   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3466   PhiNode*    result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
3467   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
3468   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3469 
3470   bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3471   Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
3472                                                   result_reg, _slow_path);
3473   Node* normal_ctl   = control();
3474   Node* no_array_ctl = result_reg->in(_slow_path);
3475 
3476   // Generate code for the slow case.  We make a call to newArray().
3477   set_control(no_array_ctl);
3478   if (!stopped()) {
3479     // Either the input type is void.class, or else the
3480     // array klass has not yet been cached.  Either the
3481     // ensuing call will throw an exception, or else it
3482     // will cache the array klass for next time.
3483     PreserveJVMState pjvms(this);
3484     CallJavaNode* slow_call = NULL;
3485     if (uninitialized) {
3486       // Generate optimized virtual call (holder class 'Unsafe' is final)
3487       slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
3488     } else {
3489       slow_call = generate_method_call_static(vmIntrinsics::_newArray);
3490     }
3491     Node* slow_result = set_results_for_java_call(slow_call);
3492     // this->control() comes from set_results_for_java_call
3493     result_reg->set_req(_slow_path, control());
3494     result_val->set_req(_slow_path, slow_result);
3495     result_io ->set_req(_slow_path, i_o());
3496     result_mem->set_req(_slow_path, reset_memory());
3497   }
3498 
3499   set_control(normal_ctl);
3500   if (!stopped()) {
3501     // Normal case:  The array type has been cached in the java.lang.Class.
3502     // The following call works fine even if the array type is polymorphic.
3503     // It could be a dynamic mix of int[], boolean[], Object[], etc.
3504     Node* obj = new_array(klass_node, count_val, 0);  // no arguments to push
3505     result_reg->init_req(_normal_path, control());
3506     result_val->init_req(_normal_path, obj);
3507     result_io ->init_req(_normal_path, i_o());
3508     result_mem->init_req(_normal_path, reset_memory());
3509 
3510     if (uninitialized) {
3511       // Mark the allocation so that zeroing is skipped
3512       AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
3513       alloc->maybe_set_complete(&_gvn);
3514     }
3515   }
3516 
3517   // Return the combined state.
3518   set_i_o(        _gvn.transform(result_io)  );
3519   set_all_memory( _gvn.transform(result_mem));
3520 
3521   C->set_has_split_ifs(true); // Has chance for split-if optimization
3522   set_result(result_reg, result_val);
3523   return true;
3524 }
3525 
3526 //----------------------inline_native_getLength--------------------------
3527 // public static native int java.lang.reflect.Array.getLength(Object array);
3528 bool LibraryCallKit::inline_native_getLength() {
3529   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3530 
3531   Node* array = null_check(argument(0));
3532   // If array is dead, only null-path is taken.
3533   if (stopped())  return true;
3534 
3535   // Deoptimize if it is a non-array.
3536   Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);
3537 
3538   if (non_array != NULL) {
3539     PreserveJVMState pjvms(this);
3540     set_control(non_array);
3541     uncommon_trap(Deoptimization::Reason_intrinsic,
3542                   Deoptimization::Action_maybe_recompile);
3543   }
3544 
3545   // If control is dead, only non-array-path is taken.
3546   if (stopped())  return true;
3547 
3548   // The works fine even if the array type is polymorphic.
3549   // It could be a dynamic mix of int[], boolean[], Object[], etc.
3550   Node* result = load_array_length(array);
3551 
3552   C->set_has_split_ifs(true);  // Has chance for split-if optimization
3553   set_result(result);
3554   return true;
3555 }
3556 
3557 //------------------------inline_array_copyOf----------------------------
3558 // public static <T,U> T[] java.util.Arrays.copyOf(     U[] original, int newLength,         Class<? extends T[]> newType);
3559 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from,      int to, Class<? extends T[]> newType);
3560 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
3561   if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;
3562 
3563   // Get the arguments.
3564   Node* original          = argument(0);
3565   Node* start             = is_copyOfRange? argument(1): intcon(0);
3566   Node* end               = is_copyOfRange? argument(2): argument(1);
3567   Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
3568 
3569   Node* newcopy = NULL;
3570 
3571   // Set the original stack and the reexecute bit for the interpreter to reexecute
3572   // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3573   { PreserveReexecuteState preexecs(this);
3574     jvms()->set_should_reexecute(true);
3575 
3576     array_type_mirror = null_check(array_type_mirror);
3577     original          = null_check(original);
3578 
3579     // Check if a null path was taken unconditionally.
3580     if (stopped())  return true;
3581 
3582     Node* orig_length = load_array_length(original);
3583 
3584     Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
3585     klass_node = null_check(klass_node);
3586 
3587     RegionNode* bailout = new RegionNode(1);
3588     record_for_igvn(bailout);
3589 
3590     // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
3591     // Bail out if that is so.
3592     Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
3593     if (not_objArray != NULL) {
3594       // Improve the klass node's type from the new optimistic assumption:
3595       ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
3596       const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3597       Node* cast = new CastPPNode(klass_node, akls);
3598       cast->init_req(0, control());
3599       klass_node = _gvn.transform(cast);
3600     }
3601 
3602     // Bail out if either start or end is negative.
3603     generate_negative_guard(start, bailout, &start);
3604     generate_negative_guard(end,   bailout, &end);
3605 
3606     Node* length = end;
3607     if (_gvn.type(start) != TypeInt::ZERO) {
3608       length = _gvn.transform(new SubINode(end, start));
3609     }
3610 
3611     // Bail out if length is negative.
3612     // Without this the new_array would throw
3613     // NegativeArraySizeException but IllegalArgumentException is what
3614     // should be thrown
3615     generate_negative_guard(length, bailout, &length);
3616 
3617     if (bailout->req() > 1) {
3618       PreserveJVMState pjvms(this);
3619       set_control(_gvn.transform(bailout));
3620       uncommon_trap(Deoptimization::Reason_intrinsic,
3621                     Deoptimization::Action_maybe_recompile);
3622     }
3623 
3624     if (!stopped()) {
3625       // How many elements will we copy from the original?
3626       // The answer is MinI(orig_length - start, length).
3627       Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
3628       Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
3629 
3630       // Generate a direct call to the right arraycopy function(s).
3631       // We know the copy is disjoint but we might not know if the
3632       // oop stores need checking.
3633       // Extreme case:  Arrays.copyOf((Integer[])x, 10, String[].class).
3634       // This will fail a store-check if x contains any non-nulls.
3635 
3636       // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
3637       // loads/stores but it is legal only if we're sure the
3638       // Arrays.copyOf would succeed. So we need all input arguments
3639       // to the copyOf to be validated, including that the copy to the
3640       // new array won't trigger an ArrayStoreException. That subtype
3641       // check can be optimized if we know something on the type of
3642       // the input array from type speculation.
3643       if (_gvn.type(klass_node)->singleton()) {
3644         ciKlass* subk   = _gvn.type(load_object_klass(original))->is_klassptr()->klass();
3645         ciKlass* superk = _gvn.type(klass_node)->is_klassptr()->klass();
3646 
3647         int test = C->static_subtype_check(superk, subk);
3648         if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
3649           const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
3650           if (t_original->speculative_type() != NULL) {
3651             original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
3652           }
3653         }
3654       }
3655 
3656       bool validated = false;
3657       // Reason_class_check rather than Reason_intrinsic because we
3658       // want to intrinsify even if this traps.
3659       if (!too_many_traps(Deoptimization::Reason_class_check)) {
3660         Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
3661 
3662         if (not_subtype_ctrl != top()) {
3663           PreserveJVMState pjvms(this);
3664           set_control(not_subtype_ctrl);
3665           uncommon_trap(Deoptimization::Reason_class_check,
3666                         Deoptimization::Action_make_not_entrant);
3667           assert(stopped(), "Should be stopped");
3668         }
3669         validated = true;
3670       }
3671 
3672       if (!stopped()) {
3673         newcopy = new_array(klass_node, length, 0);  // no arguments to push
3674 
3675         ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
3676                                                 load_object_klass(original), klass_node);
3677         if (!is_copyOfRange) {
3678           ac->set_copyof(validated);
3679         } else {
3680           ac->set_copyofrange(validated);
3681         }
3682         Node* n = _gvn.transform(ac);
3683         if (n == ac) {
3684           ac->connect_outputs(this);
3685         } else {
3686           assert(validated, "shouldn't transform if all arguments not validated");
3687           set_all_memory(n);
3688         }
3689       }
3690     }
3691   } // original reexecute is set back here
3692 
3693   C->set_has_split_ifs(true); // Has chance for split-if optimization
3694   if (!stopped()) {
3695     set_result(newcopy);
3696   }
3697   return true;
3698 }
3699 
3700 
3701 //----------------------generate_virtual_guard---------------------------
3702 // Helper for hashCode and clone.  Peeks inside the vtable to avoid a call.
3703 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
3704                                              RegionNode* slow_region) {
3705   ciMethod* method = callee();
3706   int vtable_index = method->vtable_index();
3707   assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3708          "bad index %d", vtable_index);
3709   // Get the Method* out of the appropriate vtable entry.
3710   int entry_offset  = in_bytes(Klass::vtable_start_offset()) +
3711                      vtable_index*vtableEntry::size_in_bytes() +
3712                      vtableEntry::method_offset_in_bytes();
3713   Node* entry_addr  = basic_plus_adr(obj_klass, entry_offset);
3714   Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3715 
3716   // Compare the target method with the expected method (e.g., Object.hashCode).
3717   const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
3718 
3719   Node* native_call = makecon(native_call_addr);
3720   Node* chk_native  = _gvn.transform(new CmpPNode(target_call, native_call));
3721   Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
3722 
3723   return generate_slow_guard(test_native, slow_region);
3724 }
3725 
3726 //-----------------------generate_method_call----------------------------
3727 // Use generate_method_call to make a slow-call to the real
3728 // method if the fast path fails.  An alternative would be to
3729 // use a stub like OptoRuntime::slow_arraycopy_Java.
3730 // This only works for expanding the current library call,
3731 // not another intrinsic.  (E.g., don't use this for making an
3732 // arraycopy call inside of the copyOf intrinsic.)
3733 CallJavaNode*
3734 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
3735   // When compiling the intrinsic method itself, do not use this technique.
3736   guarantee(callee() != C->method(), "cannot make slow-call to self");
3737 
3738   ciMethod* method = callee();
3739   // ensure the JVMS we have will be correct for this call
3740   guarantee(method_id == method->intrinsic_id(), "must match");
3741 
3742   const TypeFunc* tf = TypeFunc::make(method);
3743   CallJavaNode* slow_call;
3744   if (is_static) {
3745     assert(!is_virtual, "");
3746     slow_call = new CallStaticJavaNode(C, tf,
3747                            SharedRuntime::get_resolve_static_call_stub(), method);
3748   } else if (is_virtual) {
3749     null_check_receiver();
3750     int vtable_index = Method::invalid_vtable_index;
3751     if (UseInlineCaches) {
3752       // Suppress the vtable call
3753     } else {
3754       // hashCode and clone are not a miranda methods,
3755       // so the vtable index is fixed.
3756       // No need to use the linkResolver to get it.
3757        vtable_index = method->vtable_index();
3758        assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
3759               "bad index %d", vtable_index);
3760     }
3761     slow_call = new CallDynamicJavaNode(tf,
3762                           SharedRuntime::get_resolve_virtual_call_stub(),
3763                           method, vtable_index);
3764   } else {  // neither virtual nor static:  opt_virtual
3765     null_check_receiver();
3766     slow_call = new CallStaticJavaNode(C, tf,
3767                                 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
3768     slow_call->set_optimized_virtual(true);
3769   }
3770   if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
3771     // To be able to issue a direct call (optimized virtual or virtual)
3772     // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
3773     // about the method being invoked should be attached to the call site to
3774     // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
3775     slow_call->set_override_symbolic_info(true);
3776   }
3777   set_arguments_for_java_call(slow_call);
3778   set_edges_for_java_call(slow_call);
3779   return slow_call;
3780 }
3781 
3782 
3783 /**
3784  * Build special case code for calls to hashCode on an object. This call may
3785  * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
3786  * slightly different code.
3787  */
3788 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
3789   assert(is_static == callee()->is_static(), "correct intrinsic selection");
3790   assert(!(is_virtual && is_static), "either virtual, special, or static");
3791 
3792   enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
3793 
3794   RegionNode* result_reg = new RegionNode(PATH_LIMIT);
3795   PhiNode*    result_val = new PhiNode(result_reg, TypeInt::INT);
3796   PhiNode*    result_io  = new PhiNode(result_reg, Type::ABIO);
3797   PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
3798   Node* obj = NULL;
3799   if (!is_static) {
3800     // Check for hashing null object
3801     obj = null_check_receiver();
3802     if (stopped())  return true;        // unconditionally null
3803     result_reg->init_req(_null_path, top());
3804     result_val->init_req(_null_path, top());
3805   } else {
3806     // Do a null check, and return zero if null.
3807     // System.identityHashCode(null) == 0
3808     obj = argument(0);
3809     Node* null_ctl = top();
3810     obj = null_check_oop(obj, &null_ctl);
3811     result_reg->init_req(_null_path, null_ctl);
3812     result_val->init_req(_null_path, _gvn.intcon(0));
3813   }
3814 
3815   // Unconditionally null?  Then return right away.
3816   if (stopped()) {
3817     set_control( result_reg->in(_null_path));
3818     if (!stopped())
3819       set_result(result_val->in(_null_path));
3820     return true;
3821   }
3822 
3823   // We only go to the fast case code if we pass a number of guards.  The
3824   // paths which do not pass are accumulated in the slow_region.
3825   RegionNode* slow_region = new RegionNode(1);
3826   record_for_igvn(slow_region);
3827 
3828   // If this is a virtual call, we generate a funny guard.  We pull out
3829   // the vtable entry corresponding to hashCode() from the target object.
3830   // If the target method which we are calling happens to be the native
3831   // Object hashCode() method, we pass the guard.  We do not need this
3832   // guard for non-virtual calls -- the caller is known to be the native
3833   // Object hashCode().
3834   if (is_virtual) {
3835     // After null check, get the object's klass.
3836     Node* obj_klass = load_object_klass(obj);
3837     generate_virtual_guard(obj_klass, slow_region);
3838   }
3839 
3840   // Get the header out of the object, use LoadMarkNode when available
3841   Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
3842   // The control of the load must be NULL. Otherwise, the load can move before
3843   // the null check after castPP removal.
3844   Node* no_ctrl = NULL;
3845   Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
3846 
3847   // Test the header to see if it is unlocked.
3848   Node *lock_mask      = _gvn.MakeConX(markWord::lock_mask_in_place);
3849   Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
3850   Node *unlocked_val   = _gvn.MakeConX(markWord::unlocked_value);
3851   Node *chk_unlocked   = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
3852   Node *test_unlocked  = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
3853 
3854   generate_slow_guard(test_unlocked, slow_region);
3855 
3856   // Get the hash value and check to see that it has been properly assigned.
3857   // We depend on hash_mask being at most 32 bits and avoid the use of
3858   // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
3859   // vm: see markWord.hpp.
3860   Node *hash_mask      = _gvn.intcon(markWord::hash_mask);
3861   Node *hash_shift     = _gvn.intcon(markWord::hash_shift);
3862   Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
3863   // This hack lets the hash bits live anywhere in the mark object now, as long
3864   // as the shift drops the relevant bits into the low 32 bits.  Note that
3865   // Java spec says that HashCode is an int so there's no point in capturing
3866   // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
3867   hshifted_header      = ConvX2I(hshifted_header);
3868   Node *hash_val       = _gvn.transform(new AndINode(hshifted_header, hash_mask));
3869 
3870   Node *no_hash_val    = _gvn.intcon(markWord::no_hash);
3871   Node *chk_assigned   = _gvn.transform(new CmpINode( hash_val, no_hash_val));
3872   Node *test_assigned  = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
3873 
3874   generate_slow_guard(test_assigned, slow_region);
3875 
3876   Node* init_mem = reset_memory();
3877   // fill in the rest of the null path:
3878   result_io ->init_req(_null_path, i_o());
3879   result_mem->init_req(_null_path, init_mem);
3880 
3881   result_val->init_req(_fast_path, hash_val);
3882   result_reg->init_req(_fast_path, control());
3883   result_io ->init_req(_fast_path, i_o());
3884   result_mem->init_req(_fast_path, init_mem);
3885 
3886   // Generate code for the slow case.  We make a call to hashCode().
3887   set_control(_gvn.transform(slow_region));
3888   if (!stopped()) {
3889     // No need for PreserveJVMState, because we're using up the present state.
3890     set_all_memory(init_mem);
3891     vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
3892     CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
3893     Node* slow_result = set_results_for_java_call(slow_call);
3894     // this->control() comes from set_results_for_java_call
3895     result_reg->init_req(_slow_path, control());
3896     result_val->init_req(_slow_path, slow_result);
3897     result_io  ->set_req(_slow_path, i_o());
3898     result_mem ->set_req(_slow_path, reset_memory());
3899   }
3900 
3901   // Return the combined state.
3902   set_i_o(        _gvn.transform(result_io)  );
3903   set_all_memory( _gvn.transform(result_mem));
3904 
3905   set_result(result_reg, result_val);
3906   return true;
3907 }
3908 
3909 //---------------------------inline_native_getClass----------------------------
3910 // public final native Class<?> java.lang.Object.getClass();
3911 //
3912 // Build special case code for calls to getClass on an object.
3913 bool LibraryCallKit::inline_native_getClass() {
3914   Node* obj = null_check_receiver();
3915   if (stopped())  return true;
3916   set_result(load_mirror_from_klass(load_object_klass(obj)));
3917   return true;
3918 }
3919 
3920 //-----------------inline_native_Reflection_getCallerClass---------------------
3921 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
3922 //
3923 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
3924 //
3925 // NOTE: This code must perform the same logic as JVM_GetCallerClass
3926 // in that it must skip particular security frames and checks for
3927 // caller sensitive methods.
3928 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
3929 #ifndef PRODUCT
3930   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3931     tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
3932   }
3933 #endif
3934 
3935   if (!jvms()->has_method()) {
3936 #ifndef PRODUCT
3937     if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3938       tty->print_cr("  Bailing out because intrinsic was inlined at top level");
3939     }
3940 #endif
3941     return false;
3942   }
3943 
3944   // Walk back up the JVM state to find the caller at the required
3945   // depth.
3946   JVMState* caller_jvms = jvms();
3947 
3948   // Cf. JVM_GetCallerClass
3949   // NOTE: Start the loop at depth 1 because the current JVM state does
3950   // not include the Reflection.getCallerClass() frame.
3951   for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
3952     ciMethod* m = caller_jvms->method();
3953     switch (n) {
3954     case 0:
3955       fatal("current JVM state does not include the Reflection.getCallerClass frame");
3956       break;
3957     case 1:
3958       // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
3959       if (!m->caller_sensitive()) {
3960 #ifndef PRODUCT
3961         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3962           tty->print_cr("  Bailing out: CallerSensitive annotation expected at frame %d", n);
3963         }
3964 #endif
3965         return false;  // bail-out; let JVM_GetCallerClass do the work
3966       }
3967       break;
3968     default:
3969       if (!m->is_ignored_by_security_stack_walk()) {
3970         // We have reached the desired frame; return the holder class.
3971         // Acquire method holder as java.lang.Class and push as constant.
3972         ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
3973         ciInstance* caller_mirror = caller_klass->java_mirror();
3974         set_result(makecon(TypeInstPtr::make(caller_mirror)));
3975 
3976 #ifndef PRODUCT
3977         if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3978           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());
3979           tty->print_cr("  JVM state at this point:");
3980           for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
3981             ciMethod* m = jvms()->of_depth(i)->method();
3982             tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
3983           }
3984         }
3985 #endif
3986         return true;
3987       }
3988       break;
3989     }
3990   }
3991 
3992 #ifndef PRODUCT
3993   if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
3994     tty->print_cr("  Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
3995     tty->print_cr("  JVM state at this point:");
3996     for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
3997       ciMethod* m = jvms()->of_depth(i)->method();
3998       tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
3999     }
4000   }
4001 #endif
4002 
4003   return false;  // bail-out; let JVM_GetCallerClass do the work
4004 }
4005 
4006 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4007   Node* arg = argument(0);
4008   Node* result = NULL;
4009 
4010   switch (id) {
4011   case vmIntrinsics::_floatToRawIntBits:    result = new MoveF2INode(arg);  break;
4012   case vmIntrinsics::_intBitsToFloat:       result = new MoveI2FNode(arg);  break;
4013   case vmIntrinsics::_doubleToRawLongBits:  result = new MoveD2LNode(arg);  break;
4014   case vmIntrinsics::_longBitsToDouble:     result = new MoveL2DNode(arg);  break;
4015 
4016   case vmIntrinsics::_doubleToLongBits: {
4017     // two paths (plus control) merge in a wood
4018     RegionNode *r = new RegionNode(3);
4019     Node *phi = new PhiNode(r, TypeLong::LONG);
4020 
4021     Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4022     // Build the boolean node
4023     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4024 
4025     // Branch either way.
4026     // NaN case is less traveled, which makes all the difference.
4027     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4028     Node *opt_isnan = _gvn.transform(ifisnan);
4029     assert( opt_isnan->is_If(), "Expect an IfNode");
4030     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4031     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4032 
4033     set_control(iftrue);
4034 
4035     static const jlong nan_bits = CONST64(0x7ff8000000000000);
4036     Node *slow_result = longcon(nan_bits); // return NaN
4037     phi->init_req(1, _gvn.transform( slow_result ));
4038     r->init_req(1, iftrue);
4039 
4040     // Else fall through
4041     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4042     set_control(iffalse);
4043 
4044     phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4045     r->init_req(2, iffalse);
4046 
4047     // Post merge
4048     set_control(_gvn.transform(r));
4049     record_for_igvn(r);
4050 
4051     C->set_has_split_ifs(true); // Has chance for split-if optimization
4052     result = phi;
4053     assert(result->bottom_type()->isa_long(), "must be");
4054     break;
4055   }
4056 
4057   case vmIntrinsics::_floatToIntBits: {
4058     // two paths (plus control) merge in a wood
4059     RegionNode *r = new RegionNode(3);
4060     Node *phi = new PhiNode(r, TypeInt::INT);
4061 
4062     Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4063     // Build the boolean node
4064     Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4065 
4066     // Branch either way.
4067     // NaN case is less traveled, which makes all the difference.
4068     IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4069     Node *opt_isnan = _gvn.transform(ifisnan);
4070     assert( opt_isnan->is_If(), "Expect an IfNode");
4071     IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4072     Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4073 
4074     set_control(iftrue);
4075 
4076     static const jint nan_bits = 0x7fc00000;
4077     Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4078     phi->init_req(1, _gvn.transform( slow_result ));
4079     r->init_req(1, iftrue);
4080 
4081     // Else fall through
4082     Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4083     set_control(iffalse);
4084 
4085     phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4086     r->init_req(2, iffalse);
4087 
4088     // Post merge
4089     set_control(_gvn.transform(r));
4090     record_for_igvn(r);
4091 
4092     C->set_has_split_ifs(true); // Has chance for split-if optimization
4093     result = phi;
4094     assert(result->bottom_type()->isa_int(), "must be");
4095     break;
4096   }
4097 
4098   default:
4099     fatal_unexpected_iid(id);
4100     break;
4101   }
4102   set_result(_gvn.transform(result));
4103   return true;
4104 }
4105 
4106 //----------------------inline_unsafe_copyMemory-------------------------
4107 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4108 
4109 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4110   const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4111   const Type*       base_t = gvn.type(base);
4112 
4113   bool in_native = (base_t == TypePtr::NULL_PTR);
4114   bool in_heap   = !TypePtr::NULL_PTR->higher_equal(base_t);
4115   bool is_mixed  = !in_heap && !in_native;
4116 
4117   if (is_mixed) {
4118     return true; // mixed accesses can touch both on-heap and off-heap memory
4119   }
4120   if (in_heap) {
4121     bool is_prim_array = (addr_t != NULL) && (addr_t->elem() != Type::BOTTOM);
4122     if (!is_prim_array) {
4123       // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4124       // there's not enough type information available to determine proper memory slice for it.
4125       return true;
4126     }
4127   }
4128   return false;
4129 }
4130 
4131 bool LibraryCallKit::inline_unsafe_copyMemory() {
4132   if (callee()->is_static())  return false;  // caller must have the capability!
4133   null_check_receiver();  // null-check receiver
4134   if (stopped())  return true;
4135 
4136   C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".
4137 
4138   Node* src_base =         argument(1);  // type: oop
4139   Node* src_off  = ConvL2X(argument(2)); // type: long
4140   Node* dst_base =         argument(4);  // type: oop
4141   Node* dst_off  = ConvL2X(argument(5)); // type: long
4142   Node* size     = ConvL2X(argument(7)); // type: long
4143 
4144   assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4145          "fieldOffset must be byte-scaled");
4146 
4147   Node* src_addr = make_unsafe_address(src_base, src_off);
4148   Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4149 
4150   Node* thread = _gvn.transform(new ThreadLocalNode());
4151   Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4152   BasicType doing_unsafe_access_bt = T_BYTE;
4153   assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4154 
4155   // update volatile field
4156   store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4157 
4158   int flags = RC_LEAF | RC_NO_FP;
4159 
4160   const TypePtr* dst_type = TypePtr::BOTTOM;
4161 
4162   // Adjust memory effects of the runtime call based on input values.
4163   if (!has_wide_mem(_gvn, src_addr, src_base) &&
4164       !has_wide_mem(_gvn, dst_addr, dst_base)) {
4165     dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4166 
4167     const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4168     if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4169       flags |= RC_NARROW_MEM; // narrow in memory
4170     }
4171   }
4172 
4173   // Call it.  Note that the length argument is not scaled.
4174   make_runtime_call(flags,
4175                     OptoRuntime::fast_arraycopy_Type(),
4176                     StubRoutines::unsafe_arraycopy(),
4177                     "unsafe_arraycopy",
4178                     dst_type,
4179                     src_addr, dst_addr, size XTOP);
4180 
4181   store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4182 
4183   return true;
4184 }
4185 
4186 #undef XTOP
4187 
4188 //------------------------clone_coping-----------------------------------
4189 // Helper function for inline_native_clone.
4190 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4191   assert(obj_size != NULL, "");
4192   Node* raw_obj = alloc_obj->in(1);
4193   assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4194 
4195   AllocateNode* alloc = NULL;
4196   if (ReduceBulkZeroing) {
4197     // We will be completely responsible for initializing this object -
4198     // mark Initialize node as complete.
4199     alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4200     // The object was just allocated - there should be no any stores!
4201     guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4202     // Mark as complete_with_arraycopy so that on AllocateNode
4203     // expansion, we know this AllocateNode is initialized by an array
4204     // copy and a StoreStore barrier exists after the array copy.
4205     alloc->initialization()->set_complete_with_arraycopy();
4206   }
4207 
4208   Node* size = _gvn.transform(obj_size);
4209   access_clone(obj, alloc_obj, size, is_array);
4210 
4211   // Do not let reads from the cloned object float above the arraycopy.
4212   if (alloc != NULL) {
4213     // Do not let stores that initialize this object be reordered with
4214     // a subsequent store that would make this object accessible by
4215     // other threads.
4216     // Record what AllocateNode this StoreStore protects so that
4217     // escape analysis can go from the MemBarStoreStoreNode to the
4218     // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4219     // based on the escape status of the AllocateNode.
4220     insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4221   } else {
4222     insert_mem_bar(Op_MemBarCPUOrder);
4223   }
4224 }
4225 
4226 //------------------------inline_native_clone----------------------------
4227 // protected native Object java.lang.Object.clone();
4228 //
4229 // Here are the simple edge cases:
4230 //  null receiver => normal trap
4231 //  virtual and clone was overridden => slow path to out-of-line clone
4232 //  not cloneable or finalizer => slow path to out-of-line Object.clone
4233 //
4234 // The general case has two steps, allocation and copying.
4235 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
4236 //
4237 // Copying also has two cases, oop arrays and everything else.
4238 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
4239 // Everything else uses the tight inline loop supplied by CopyArrayNode.
4240 //
4241 // These steps fold up nicely if and when the cloned object's klass
4242 // can be sharply typed as an object array, a type array, or an instance.
4243 //
4244 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4245   PhiNode* result_val;
4246 
4247   // Set the reexecute bit for the interpreter to reexecute
4248   // the bytecode that invokes Object.clone if deoptimization happens.
4249   { PreserveReexecuteState preexecs(this);
4250     jvms()->set_should_reexecute(true);
4251 
4252     Node* obj = null_check_receiver();
4253     if (stopped())  return true;
4254 
4255     const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
4256 
4257     // If we are going to clone an instance, we need its exact type to
4258     // know the number and types of fields to convert the clone to
4259     // loads/stores. Maybe a speculative type can help us.
4260     if (!obj_type->klass_is_exact() &&
4261         obj_type->speculative_type() != NULL &&
4262         obj_type->speculative_type()->is_instance_klass()) {
4263       ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
4264       if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
4265           !spec_ik->has_injected_fields()) {
4266         ciKlass* k = obj_type->klass();
4267         if (!k->is_instance_klass() ||
4268             k->as_instance_klass()->is_interface() ||
4269             k->as_instance_klass()->has_subklass()) {
4270           obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
4271         }
4272       }
4273     }
4274 
4275     // Conservatively insert a memory barrier on all memory slices.
4276     // Do not let writes into the original float below the clone.
4277     insert_mem_bar(Op_MemBarCPUOrder);
4278 
4279     // paths into result_reg:
4280     enum {
4281       _slow_path = 1,     // out-of-line call to clone method (virtual or not)
4282       _objArray_path,     // plain array allocation, plus arrayof_oop_arraycopy
4283       _array_path,        // plain array allocation, plus arrayof_long_arraycopy
4284       _instance_path,     // plain instance allocation, plus arrayof_long_arraycopy
4285       PATH_LIMIT
4286     };
4287     RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4288     result_val             = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4289     PhiNode*    result_i_o = new PhiNode(result_reg, Type::ABIO);
4290     PhiNode*    result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4291     record_for_igvn(result_reg);
4292 
4293     Node* obj_klass = load_object_klass(obj);
4294     Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
4295     if (array_ctl != NULL) {
4296       // It's an array.
4297       PreserveJVMState pjvms(this);
4298       set_control(array_ctl);
4299       Node* obj_length = load_array_length(obj);
4300       Node* obj_size  = NULL;
4301       Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
4302 
4303       BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4304       if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
4305         // If it is an oop array, it requires very special treatment,
4306         // because gc barriers are required when accessing the array.
4307         Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
4308         if (is_obja != NULL) {
4309           PreserveJVMState pjvms2(this);
4310           set_control(is_obja);
4311           // Generate a direct call to the right arraycopy function(s).
4312           // Clones are always tightly coupled.
4313           ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
4314           ac->set_clone_oop_array();
4315           Node* n = _gvn.transform(ac);
4316           assert(n == ac, "cannot disappear");
4317           ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
4318 
4319           result_reg->init_req(_objArray_path, control());
4320           result_val->init_req(_objArray_path, alloc_obj);
4321           result_i_o ->set_req(_objArray_path, i_o());
4322           result_mem ->set_req(_objArray_path, reset_memory());
4323         }
4324       }
4325       // Otherwise, there are no barriers to worry about.
4326       // (We can dispense with card marks if we know the allocation
4327       //  comes out of eden (TLAB)...  In fact, ReduceInitialCardMarks
4328       //  causes the non-eden paths to take compensating steps to
4329       //  simulate a fresh allocation, so that no further
4330       //  card marks are required in compiled code to initialize
4331       //  the object.)
4332 
4333       if (!stopped()) {
4334         copy_to_clone(obj, alloc_obj, obj_size, true);
4335 
4336         // Present the results of the copy.
4337         result_reg->init_req(_array_path, control());
4338         result_val->init_req(_array_path, alloc_obj);
4339         result_i_o ->set_req(_array_path, i_o());
4340         result_mem ->set_req(_array_path, reset_memory());
4341       }
4342     }
4343 
4344     // We only go to the instance fast case code if we pass a number of guards.
4345     // The paths which do not pass are accumulated in the slow_region.
4346     RegionNode* slow_region = new RegionNode(1);
4347     record_for_igvn(slow_region);
4348     if (!stopped()) {
4349       // It's an instance (we did array above).  Make the slow-path tests.
4350       // If this is a virtual call, we generate a funny guard.  We grab
4351       // the vtable entry corresponding to clone() from the target object.
4352       // If the target method which we are calling happens to be the
4353       // Object clone() method, we pass the guard.  We do not need this
4354       // guard for non-virtual calls; the caller is known to be the native
4355       // Object clone().
4356       if (is_virtual) {
4357         generate_virtual_guard(obj_klass, slow_region);
4358       }
4359 
4360       // The object must be easily cloneable and must not have a finalizer.
4361       // Both of these conditions may be checked in a single test.
4362       // We could optimize the test further, but we don't care.
4363       generate_access_flags_guard(obj_klass,
4364                                   // Test both conditions:
4365                                   JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
4366                                   // Must be cloneable but not finalizer:
4367                                   JVM_ACC_IS_CLONEABLE_FAST,
4368                                   slow_region);
4369     }
4370 
4371     if (!stopped()) {
4372       // It's an instance, and it passed the slow-path tests.
4373       PreserveJVMState pjvms(this);
4374       Node* obj_size  = NULL;
4375       // Need to deoptimize on exception from allocation since Object.clone intrinsic
4376       // is reexecuted if deoptimization occurs and there could be problems when merging
4377       // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
4378       Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size, /*deoptimize_on_exception=*/true);
4379 
4380       copy_to_clone(obj, alloc_obj, obj_size, false);
4381 
4382       // Present the results of the slow call.
4383       result_reg->init_req(_instance_path, control());
4384       result_val->init_req(_instance_path, alloc_obj);
4385       result_i_o ->set_req(_instance_path, i_o());
4386       result_mem ->set_req(_instance_path, reset_memory());
4387     }
4388 
4389     // Generate code for the slow case.  We make a call to clone().
4390     set_control(_gvn.transform(slow_region));
4391     if (!stopped()) {
4392       PreserveJVMState pjvms(this);
4393       CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
4394       // We need to deoptimize on exception (see comment above)
4395       Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
4396       // this->control() comes from set_results_for_java_call
4397       result_reg->init_req(_slow_path, control());
4398       result_val->init_req(_slow_path, slow_result);
4399       result_i_o ->set_req(_slow_path, i_o());
4400       result_mem ->set_req(_slow_path, reset_memory());
4401     }
4402 
4403     // Return the combined state.
4404     set_control(    _gvn.transform(result_reg));
4405     set_i_o(        _gvn.transform(result_i_o));
4406     set_all_memory( _gvn.transform(result_mem));
4407   } // original reexecute is set back here
4408 
4409   set_result(_gvn.transform(result_val));
4410   return true;
4411 }
4412 
4413 // If we have a tightly coupled allocation, the arraycopy may take care
4414 // of the array initialization. If one of the guards we insert between
4415 // the allocation and the arraycopy causes a deoptimization, an
4416 // unitialized array will escape the compiled method. To prevent that
4417 // we set the JVM state for uncommon traps between the allocation and
4418 // the arraycopy to the state before the allocation so, in case of
4419 // deoptimization, we'll reexecute the allocation and the
4420 // initialization.
4421 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
4422   if (alloc != NULL) {
4423     ciMethod* trap_method = alloc->jvms()->method();
4424     int trap_bci = alloc->jvms()->bci();
4425 
4426     if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4427         !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
4428       // Make sure there's no store between the allocation and the
4429       // arraycopy otherwise visible side effects could be rexecuted
4430       // in case of deoptimization and cause incorrect execution.
4431       bool no_interfering_store = true;
4432       Node* mem = alloc->in(TypeFunc::Memory);
4433       if (mem->is_MergeMem()) {
4434         for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
4435           Node* n = mms.memory();
4436           if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4437             assert(n->is_Store(), "what else?");
4438             no_interfering_store = false;
4439             break;
4440           }
4441         }
4442       } else {
4443         for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
4444           Node* n = mms.memory();
4445           if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
4446             assert(n->is_Store(), "what else?");
4447             no_interfering_store = false;
4448             break;
4449           }
4450         }
4451       }
4452 
4453       if (no_interfering_store) {
4454         JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
4455         uint size = alloc->req();
4456         SafePointNode* sfpt = new SafePointNode(size, old_jvms);
4457         old_jvms->set_map(sfpt);
4458         for (uint i = 0; i < size; i++) {
4459           sfpt->init_req(i, alloc->in(i));
4460         }
4461         // re-push array length for deoptimization
4462         sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
4463         old_jvms->set_sp(old_jvms->sp()+1);
4464         old_jvms->set_monoff(old_jvms->monoff()+1);
4465         old_jvms->set_scloff(old_jvms->scloff()+1);
4466         old_jvms->set_endoff(old_jvms->endoff()+1);
4467         old_jvms->set_should_reexecute(true);
4468 
4469         sfpt->set_i_o(map()->i_o());
4470         sfpt->set_memory(map()->memory());
4471         sfpt->set_control(map()->control());
4472 
4473         JVMState* saved_jvms = jvms();
4474         saved_reexecute_sp = _reexecute_sp;
4475 
4476         set_jvms(sfpt->jvms());
4477         _reexecute_sp = jvms()->sp();
4478 
4479         return saved_jvms;
4480       }
4481     }
4482   }
4483   return NULL;
4484 }
4485 
4486 // In case of a deoptimization, we restart execution at the
4487 // allocation, allocating a new array. We would leave an uninitialized
4488 // array in the heap that GCs wouldn't expect. Move the allocation
4489 // after the traps so we don't allocate the array if we
4490 // deoptimize. This is possible because tightly_coupled_allocation()
4491 // guarantees there's no observer of the allocated array at this point
4492 // and the control flow is simple enough.
4493 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms,
4494                                                     int saved_reexecute_sp, uint new_idx) {
4495   if (saved_jvms != NULL && !stopped()) {
4496     assert(alloc != NULL, "only with a tightly coupled allocation");
4497     // restore JVM state to the state at the arraycopy
4498     saved_jvms->map()->set_control(map()->control());
4499     assert(saved_jvms->map()->memory() == map()->memory(), "memory state changed?");
4500     assert(saved_jvms->map()->i_o() == map()->i_o(), "IO state changed?");
4501     // If we've improved the types of some nodes (null check) while
4502     // emitting the guards, propagate them to the current state
4503     map()->replaced_nodes().apply(saved_jvms->map(), new_idx);
4504     set_jvms(saved_jvms);
4505     _reexecute_sp = saved_reexecute_sp;
4506 
4507     // Remove the allocation from above the guards
4508     CallProjections callprojs;
4509     alloc->extract_projections(&callprojs, true);
4510     InitializeNode* init = alloc->initialization();
4511     Node* alloc_mem = alloc->in(TypeFunc::Memory);
4512     C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
4513     C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
4514 
4515     // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
4516     // the allocation (i.e. is only valid if the allocation succeeds):
4517     // 1) replace CastIINode with AllocateArrayNode's length here
4518     // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
4519     //
4520     // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
4521     // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
4522     Node* init_control = init->proj_out(TypeFunc::Control);
4523     Node* alloc_length = alloc->Ideal_length();
4524 #ifdef ASSERT
4525     Node* prev_cast = NULL;
4526 #endif
4527     for (uint i = 0; i < init_control->outcnt(); i++) {
4528       Node* init_out = init_control->raw_out(i);
4529       if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
4530 #ifdef ASSERT
4531         if (prev_cast == NULL) {
4532           prev_cast = init_out;
4533         } else {
4534           if (prev_cast->cmp(*init_out) == false) {
4535             prev_cast->dump();
4536             init_out->dump();
4537             assert(false, "not equal CastIINode");
4538           }
4539         }
4540 #endif
4541         C->gvn_replace_by(init_out, alloc_length);
4542       }
4543     }
4544     C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
4545 
4546     // move the allocation here (after the guards)
4547     _gvn.hash_delete(alloc);
4548     alloc->set_req(TypeFunc::Control, control());
4549     alloc->set_req(TypeFunc::I_O, i_o());
4550     Node *mem = reset_memory();
4551     set_all_memory(mem);
4552     alloc->set_req(TypeFunc::Memory, mem);
4553     set_control(init->proj_out_or_null(TypeFunc::Control));
4554     set_i_o(callprojs.fallthrough_ioproj);
4555 
4556     // Update memory as done in GraphKit::set_output_for_allocation()
4557     const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
4558     const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
4559     if (ary_type->isa_aryptr() && length_type != NULL) {
4560       ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
4561     }
4562     const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
4563     int            elemidx  = C->get_alias_index(telemref);
4564     set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
4565     set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
4566 
4567     Node* allocx = _gvn.transform(alloc);
4568     assert(allocx == alloc, "where has the allocation gone?");
4569     assert(dest->is_CheckCastPP(), "not an allocation result?");
4570 
4571     _gvn.hash_delete(dest);
4572     dest->set_req(0, control());
4573     Node* destx = _gvn.transform(dest);
4574     assert(destx == dest, "where has the allocation result gone?");
4575 
4576     array_ideal_length(alloc, ary_type, true);
4577   }
4578 }
4579 
4580 
4581 //------------------------------inline_arraycopy-----------------------
4582 // public static native void java.lang.System.arraycopy(Object src,  int  srcPos,
4583 //                                                      Object dest, int destPos,
4584 //                                                      int length);
4585 bool LibraryCallKit::inline_arraycopy() {
4586   // Get the arguments.
4587   Node* src         = argument(0);  // type: oop
4588   Node* src_offset  = argument(1);  // type: int
4589   Node* dest        = argument(2);  // type: oop
4590   Node* dest_offset = argument(3);  // type: int
4591   Node* length      = argument(4);  // type: int
4592 
4593   uint new_idx = C->unique();
4594 
4595   // Check for allocation before we add nodes that would confuse
4596   // tightly_coupled_allocation()
4597   AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
4598 
4599   int saved_reexecute_sp = -1;
4600   JVMState* saved_jvms = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
4601   // See arraycopy_restore_alloc_state() comment
4602   // if alloc == NULL we don't have to worry about a tightly coupled allocation so we can emit all needed guards
4603   // if saved_jvms != NULL (then alloc != NULL) then we can handle guards and a tightly coupled allocation
4604   // if saved_jvms == NULL and alloc != NULL, we can't emit any guards
4605   bool can_emit_guards = (alloc == NULL || saved_jvms != NULL);
4606 
4607   // The following tests must be performed
4608   // (1) src and dest are arrays.
4609   // (2) src and dest arrays must have elements of the same BasicType
4610   // (3) src and dest must not be null.
4611   // (4) src_offset must not be negative.
4612   // (5) dest_offset must not be negative.
4613   // (6) length must not be negative.
4614   // (7) src_offset + length must not exceed length of src.
4615   // (8) dest_offset + length must not exceed length of dest.
4616   // (9) each element of an oop array must be assignable
4617 
4618   // (3) src and dest must not be null.
4619   // always do this here because we need the JVM state for uncommon traps
4620   Node* null_ctl = top();
4621   src  = saved_jvms != NULL ? null_check_oop(src, &null_ctl, true, true) : null_check(src,  T_ARRAY);
4622   assert(null_ctl->is_top(), "no null control here");
4623   dest = null_check(dest, T_ARRAY);
4624 
4625   if (!can_emit_guards) {
4626     // if saved_jvms == NULL and alloc != NULL, we don't emit any
4627     // guards but the arraycopy node could still take advantage of a
4628     // tightly allocated allocation. tightly_coupled_allocation() is
4629     // called again to make sure it takes the null check above into
4630     // account: the null check is mandatory and if it caused an
4631     // uncommon trap to be emitted then the allocation can't be
4632     // considered tightly coupled in this context.
4633     alloc = tightly_coupled_allocation(dest);
4634   }
4635 
4636   bool validated = false;
4637 
4638   const Type* src_type  = _gvn.type(src);
4639   const Type* dest_type = _gvn.type(dest);
4640   const TypeAryPtr* top_src  = src_type->isa_aryptr();
4641   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4642 
4643   // Do we have the type of src?
4644   bool has_src = (top_src != NULL && top_src->klass() != NULL);
4645   // Do we have the type of dest?
4646   bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4647   // Is the type for src from speculation?
4648   bool src_spec = false;
4649   // Is the type for dest from speculation?
4650   bool dest_spec = false;
4651 
4652   if ((!has_src || !has_dest) && can_emit_guards) {
4653     // We don't have sufficient type information, let's see if
4654     // speculative types can help. We need to have types for both src
4655     // and dest so that it pays off.
4656 
4657     // Do we already have or could we have type information for src
4658     bool could_have_src = has_src;
4659     // Do we already have or could we have type information for dest
4660     bool could_have_dest = has_dest;
4661 
4662     ciKlass* src_k = NULL;
4663     if (!has_src) {
4664       src_k = src_type->speculative_type_not_null();
4665       if (src_k != NULL && src_k->is_array_klass()) {
4666         could_have_src = true;
4667       }
4668     }
4669 
4670     ciKlass* dest_k = NULL;
4671     if (!has_dest) {
4672       dest_k = dest_type->speculative_type_not_null();
4673       if (dest_k != NULL && dest_k->is_array_klass()) {
4674         could_have_dest = true;
4675       }
4676     }
4677 
4678     if (could_have_src && could_have_dest) {
4679       // This is going to pay off so emit the required guards
4680       if (!has_src) {
4681         src = maybe_cast_profiled_obj(src, src_k, true);
4682         src_type  = _gvn.type(src);
4683         top_src  = src_type->isa_aryptr();
4684         has_src = (top_src != NULL && top_src->klass() != NULL);
4685         src_spec = true;
4686       }
4687       if (!has_dest) {
4688         dest = maybe_cast_profiled_obj(dest, dest_k, true);
4689         dest_type  = _gvn.type(dest);
4690         top_dest  = dest_type->isa_aryptr();
4691         has_dest = (top_dest != NULL && top_dest->klass() != NULL);
4692         dest_spec = true;
4693       }
4694     }
4695   }
4696 
4697   if (has_src && has_dest && can_emit_guards) {
4698     BasicType src_elem  = top_src->klass()->as_array_klass()->element_type()->basic_type();
4699     BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
4700     if (is_reference_type(src_elem))   src_elem  = T_OBJECT;
4701     if (is_reference_type(dest_elem))  dest_elem = T_OBJECT;
4702 
4703     if (src_elem == dest_elem && src_elem == T_OBJECT) {
4704       // If both arrays are object arrays then having the exact types
4705       // for both will remove the need for a subtype check at runtime
4706       // before the call and may make it possible to pick a faster copy
4707       // routine (without a subtype check on every element)
4708       // Do we have the exact type of src?
4709       bool could_have_src = src_spec;
4710       // Do we have the exact type of dest?
4711       bool could_have_dest = dest_spec;
4712       ciKlass* src_k = top_src->klass();
4713       ciKlass* dest_k = top_dest->klass();
4714       if (!src_spec) {
4715         src_k = src_type->speculative_type_not_null();
4716         if (src_k != NULL && src_k->is_array_klass()) {
4717           could_have_src = true;
4718         }
4719       }
4720       if (!dest_spec) {
4721         dest_k = dest_type->speculative_type_not_null();
4722         if (dest_k != NULL && dest_k->is_array_klass()) {
4723           could_have_dest = true;
4724         }
4725       }
4726       if (could_have_src && could_have_dest) {
4727         // If we can have both exact types, emit the missing guards
4728         if (could_have_src && !src_spec) {
4729           src = maybe_cast_profiled_obj(src, src_k, true);
4730         }
4731         if (could_have_dest && !dest_spec) {
4732           dest = maybe_cast_profiled_obj(dest, dest_k, true);
4733         }
4734       }
4735     }
4736   }
4737 
4738   ciMethod* trap_method = method();
4739   int trap_bci = bci();
4740   if (saved_jvms != NULL) {
4741     trap_method = alloc->jvms()->method();
4742     trap_bci = alloc->jvms()->bci();
4743   }
4744 
4745   bool negative_length_guard_generated = false;
4746 
4747   if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
4748       can_emit_guards &&
4749       !src->is_top() && !dest->is_top()) {
4750     // validate arguments: enables transformation the ArrayCopyNode
4751     validated = true;
4752 
4753     RegionNode* slow_region = new RegionNode(1);
4754     record_for_igvn(slow_region);
4755 
4756     // (1) src and dest are arrays.
4757     generate_non_array_guard(load_object_klass(src), slow_region);
4758     generate_non_array_guard(load_object_klass(dest), slow_region);
4759 
4760     // (2) src and dest arrays must have elements of the same BasicType
4761     // done at macro expansion or at Ideal transformation time
4762 
4763     // (4) src_offset must not be negative.
4764     generate_negative_guard(src_offset, slow_region);
4765 
4766     // (5) dest_offset must not be negative.
4767     generate_negative_guard(dest_offset, slow_region);
4768 
4769     // (7) src_offset + length must not exceed length of src.
4770     generate_limit_guard(src_offset, length,
4771                          load_array_length(src),
4772                          slow_region);
4773 
4774     // (8) dest_offset + length must not exceed length of dest.
4775     generate_limit_guard(dest_offset, length,
4776                          load_array_length(dest),
4777                          slow_region);
4778 
4779     // (6) length must not be negative.
4780     // This is also checked in generate_arraycopy() during macro expansion, but
4781     // we also have to check it here for the case where the ArrayCopyNode will
4782     // be eliminated by Escape Analysis.
4783     if (EliminateAllocations) {
4784       generate_negative_guard(length, slow_region);
4785       negative_length_guard_generated = true;
4786     }
4787 
4788     // (9) each element of an oop array must be assignable
4789     Node* dest_klass = load_object_klass(dest);
4790     if (src != dest) {
4791       Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
4792 
4793       if (not_subtype_ctrl != top()) {
4794         PreserveJVMState pjvms(this);
4795         set_control(not_subtype_ctrl);
4796         uncommon_trap(Deoptimization::Reason_intrinsic,
4797                       Deoptimization::Action_make_not_entrant);
4798         assert(stopped(), "Should be stopped");
4799       }
4800     }
4801     {
4802       PreserveJVMState pjvms(this);
4803       set_control(_gvn.transform(slow_region));
4804       uncommon_trap(Deoptimization::Reason_intrinsic,
4805                     Deoptimization::Action_make_not_entrant);
4806       assert(stopped(), "Should be stopped");
4807     }
4808 
4809     const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
4810     const Type *toop = TypeOopPtr::make_from_klass(dest_klass_t->klass());
4811     src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
4812   }
4813 
4814   arraycopy_move_allocation_here(alloc, dest, saved_jvms, saved_reexecute_sp, new_idx);
4815 
4816   if (stopped()) {
4817     return true;
4818   }
4819 
4820   ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != NULL, negative_length_guard_generated,
4821                                           // Create LoadRange and LoadKlass nodes for use during macro expansion here
4822                                           // so the compiler has a chance to eliminate them: during macro expansion,
4823                                           // we have to set their control (CastPP nodes are eliminated).
4824                                           load_object_klass(src), load_object_klass(dest),
4825                                           load_array_length(src), load_array_length(dest));
4826 
4827   ac->set_arraycopy(validated);
4828 
4829   Node* n = _gvn.transform(ac);
4830   if (n == ac) {
4831     ac->connect_outputs(this);
4832   } else {
4833     assert(validated, "shouldn't transform if all arguments not validated");
4834     set_all_memory(n);
4835   }
4836   clear_upper_avx();
4837 
4838 
4839   return true;
4840 }
4841 
4842 
4843 // Helper function which determines if an arraycopy immediately follows
4844 // an allocation, with no intervening tests or other escapes for the object.
4845 AllocateArrayNode*
4846 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
4847   if (stopped())             return NULL;  // no fast path
4848   if (C->AliasLevel() == 0)  return NULL;  // no MergeMems around
4849 
4850   AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
4851   if (alloc == NULL)  return NULL;
4852 
4853   Node* rawmem = memory(Compile::AliasIdxRaw);
4854   // Is the allocation's memory state untouched?
4855   if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
4856     // Bail out if there have been raw-memory effects since the allocation.
4857     // (Example:  There might have been a call or safepoint.)
4858     return NULL;
4859   }
4860   rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
4861   if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
4862     return NULL;
4863   }
4864 
4865   // There must be no unexpected observers of this allocation.
4866   for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
4867     Node* obs = ptr->fast_out(i);
4868     if (obs != this->map()) {
4869       return NULL;
4870     }
4871   }
4872 
4873   // This arraycopy must unconditionally follow the allocation of the ptr.
4874   Node* alloc_ctl = ptr->in(0);
4875   Node* ctl = control();
4876   while (ctl != alloc_ctl) {
4877     // There may be guards which feed into the slow_region.
4878     // Any other control flow means that we might not get a chance
4879     // to finish initializing the allocated object.
4880     if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
4881       IfNode* iff = ctl->in(0)->as_If();
4882       Node* not_ctl = iff->proj_out_or_null(1 - ctl->as_Proj()->_con);
4883       assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
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 or ASCII
4914 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
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   const Type* src_type = src->Value(&_gvn);
4927   const Type* dst_type = dst->Value(&_gvn);
4928   const TypeAryPtr* top_src = src_type->isa_aryptr();
4929   const TypeAryPtr* top_dest = dst_type->isa_aryptr();
4930   if (top_src  == NULL || top_src->klass()  == NULL ||
4931       top_dest == NULL || top_dest->klass() == NULL) {
4932     // failed array check
4933     return false;
4934   }
4935 
4936   // Figure out the size and type of the elements we will be copying.
4937   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4938   BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4939   if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
4940     return false;
4941   }
4942 
4943   Node* src_start = array_element_address(src, src_offset, T_CHAR);
4944   Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
4945   // 'src_start' points to src array + scaled offset
4946   // 'dst_start' points to dst array + scaled offset
4947 
4948   const TypeAryPtr* mtype = TypeAryPtr::BYTES;
4949   Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
4950   enc = _gvn.transform(enc);
4951   Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
4952   set_memory(res_mem, mtype);
4953   set_result(enc);
4954   clear_upper_avx();
4955 
4956   return true;
4957 }
4958 
4959 //-------------inline_multiplyToLen-----------------------------------
4960 bool LibraryCallKit::inline_multiplyToLen() {
4961   assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
4962 
4963   address stubAddr = StubRoutines::multiplyToLen();
4964   if (stubAddr == NULL) {
4965     return false; // Intrinsic's stub is not implemented on this platform
4966   }
4967   const char* stubName = "multiplyToLen";
4968 
4969   assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
4970 
4971   // no receiver because it is a static method
4972   Node* x    = argument(0);
4973   Node* xlen = argument(1);
4974   Node* y    = argument(2);
4975   Node* ylen = argument(3);
4976   Node* z    = argument(4);
4977 
4978   x = must_be_not_null(x, true);
4979   y = must_be_not_null(y, true);
4980 
4981   const Type* x_type = x->Value(&_gvn);
4982   const Type* y_type = y->Value(&_gvn);
4983   const TypeAryPtr* top_x = x_type->isa_aryptr();
4984   const TypeAryPtr* top_y = y_type->isa_aryptr();
4985   if (top_x  == NULL || top_x->klass()  == NULL ||
4986       top_y == NULL || top_y->klass() == NULL) {
4987     // failed array check
4988     return false;
4989   }
4990 
4991   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4992   BasicType y_elem = y_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
4993   if (x_elem != T_INT || y_elem != T_INT) {
4994     return false;
4995   }
4996 
4997   // Set the original stack and the reexecute bit for the interpreter to reexecute
4998   // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
4999   // on the return from z array allocation in runtime.
5000   { PreserveReexecuteState preexecs(this);
5001     jvms()->set_should_reexecute(true);
5002 
5003     Node* x_start = array_element_address(x, intcon(0), x_elem);
5004     Node* y_start = array_element_address(y, intcon(0), y_elem);
5005     // 'x_start' points to x array + scaled xlen
5006     // 'y_start' points to y array + scaled ylen
5007 
5008     // Allocate the result array
5009     Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5010     ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5011     Node* klass_node = makecon(TypeKlassPtr::make(klass));
5012 
5013     IdealKit ideal(this);
5014 
5015 #define __ ideal.
5016      Node* one = __ ConI(1);
5017      Node* zero = __ ConI(0);
5018      IdealVariable need_alloc(ideal), z_alloc(ideal);  __ declarations_done();
5019      __ set(need_alloc, zero);
5020      __ set(z_alloc, z);
5021      __ if_then(z, BoolTest::eq, null()); {
5022        __ increment (need_alloc, one);
5023      } __ else_(); {
5024        // Update graphKit memory and control from IdealKit.
5025        sync_kit(ideal);
5026        Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
5027        cast->init_req(0, control());
5028        _gvn.set_type(cast, cast->bottom_type());
5029        C->record_for_igvn(cast);
5030 
5031        Node* zlen_arg = load_array_length(cast);
5032        // Update IdealKit memory and control from graphKit.
5033        __ sync_kit(this);
5034        __ if_then(zlen_arg, BoolTest::lt, zlen); {
5035          __ increment (need_alloc, one);
5036        } __ end_if();
5037      } __ end_if();
5038 
5039      __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5040        // Update graphKit memory and control from IdealKit.
5041        sync_kit(ideal);
5042        Node * narr = new_array(klass_node, zlen, 1);
5043        // Update IdealKit memory and control from graphKit.
5044        __ sync_kit(this);
5045        __ set(z_alloc, narr);
5046      } __ end_if();
5047 
5048      sync_kit(ideal);
5049      z = __ value(z_alloc);
5050      // Can't use TypeAryPtr::INTS which uses Bottom offset.
5051      _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5052      // Final sync IdealKit and GraphKit.
5053      final_sync(ideal);
5054 #undef __
5055 
5056     Node* z_start = array_element_address(z, intcon(0), T_INT);
5057 
5058     Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5059                                    OptoRuntime::multiplyToLen_Type(),
5060                                    stubAddr, stubName, TypePtr::BOTTOM,
5061                                    x_start, xlen, y_start, ylen, z_start, zlen);
5062   } // original reexecute is set back here
5063 
5064   C->set_has_split_ifs(true); // Has chance for split-if optimization
5065   set_result(z);
5066   return true;
5067 }
5068 
5069 //-------------inline_squareToLen------------------------------------
5070 bool LibraryCallKit::inline_squareToLen() {
5071   assert(UseSquareToLenIntrinsic, "not implemented on this platform");
5072 
5073   address stubAddr = StubRoutines::squareToLen();
5074   if (stubAddr == NULL) {
5075     return false; // Intrinsic's stub is not implemented on this platform
5076   }
5077   const char* stubName = "squareToLen";
5078 
5079   assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
5080 
5081   Node* x    = argument(0);
5082   Node* len  = argument(1);
5083   Node* z    = argument(2);
5084   Node* zlen = argument(3);
5085 
5086   x = must_be_not_null(x, true);
5087   z = must_be_not_null(z, true);
5088 
5089   const Type* x_type = x->Value(&_gvn);
5090   const Type* z_type = z->Value(&_gvn);
5091   const TypeAryPtr* top_x = x_type->isa_aryptr();
5092   const TypeAryPtr* top_z = z_type->isa_aryptr();
5093   if (top_x  == NULL || top_x->klass()  == NULL ||
5094       top_z  == NULL || top_z->klass()  == NULL) {
5095     // failed array check
5096     return false;
5097   }
5098 
5099   BasicType x_elem = x_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5100   BasicType z_elem = z_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5101   if (x_elem != T_INT || z_elem != T_INT) {
5102     return false;
5103   }
5104 
5105 
5106   Node* x_start = array_element_address(x, intcon(0), x_elem);
5107   Node* z_start = array_element_address(z, intcon(0), z_elem);
5108 
5109   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5110                                   OptoRuntime::squareToLen_Type(),
5111                                   stubAddr, stubName, TypePtr::BOTTOM,
5112                                   x_start, len, z_start, zlen);
5113 
5114   set_result(z);
5115   return true;
5116 }
5117 
5118 //-------------inline_mulAdd------------------------------------------
5119 bool LibraryCallKit::inline_mulAdd() {
5120   assert(UseMulAddIntrinsic, "not implemented on this platform");
5121 
5122   address stubAddr = StubRoutines::mulAdd();
5123   if (stubAddr == NULL) {
5124     return false; // Intrinsic's stub is not implemented on this platform
5125   }
5126   const char* stubName = "mulAdd";
5127 
5128   assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
5129 
5130   Node* out      = argument(0);
5131   Node* in       = argument(1);
5132   Node* offset   = argument(2);
5133   Node* len      = argument(3);
5134   Node* k        = argument(4);
5135 
5136   out = must_be_not_null(out, true);
5137 
5138   const Type* out_type = out->Value(&_gvn);
5139   const Type* in_type = in->Value(&_gvn);
5140   const TypeAryPtr* top_out = out_type->isa_aryptr();
5141   const TypeAryPtr* top_in = in_type->isa_aryptr();
5142   if (top_out  == NULL || top_out->klass()  == NULL ||
5143       top_in == NULL || top_in->klass() == NULL) {
5144     // failed array check
5145     return false;
5146   }
5147 
5148   BasicType out_elem = out_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5149   BasicType in_elem = in_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5150   if (out_elem != T_INT || in_elem != T_INT) {
5151     return false;
5152   }
5153 
5154   Node* outlen = load_array_length(out);
5155   Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
5156   Node* out_start = array_element_address(out, intcon(0), out_elem);
5157   Node* in_start = array_element_address(in, intcon(0), in_elem);
5158 
5159   Node*  call = make_runtime_call(RC_LEAF|RC_NO_FP,
5160                                   OptoRuntime::mulAdd_Type(),
5161                                   stubAddr, stubName, TypePtr::BOTTOM,
5162                                   out_start,in_start, new_offset, len, k);
5163   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5164   set_result(result);
5165   return true;
5166 }
5167 
5168 //-------------inline_montgomeryMultiply-----------------------------------
5169 bool LibraryCallKit::inline_montgomeryMultiply() {
5170   address stubAddr = StubRoutines::montgomeryMultiply();
5171   if (stubAddr == NULL) {
5172     return false; // Intrinsic's stub is not implemented on this platform
5173   }
5174 
5175   assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
5176   const char* stubName = "montgomery_multiply";
5177 
5178   assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
5179 
5180   Node* a    = argument(0);
5181   Node* b    = argument(1);
5182   Node* n    = argument(2);
5183   Node* len  = argument(3);
5184   Node* inv  = argument(4);
5185   Node* m    = argument(6);
5186 
5187   const Type* a_type = a->Value(&_gvn);
5188   const TypeAryPtr* top_a = a_type->isa_aryptr();
5189   const Type* b_type = b->Value(&_gvn);
5190   const TypeAryPtr* top_b = b_type->isa_aryptr();
5191   const Type* n_type = a->Value(&_gvn);
5192   const TypeAryPtr* top_n = n_type->isa_aryptr();
5193   const Type* m_type = a->Value(&_gvn);
5194   const TypeAryPtr* top_m = m_type->isa_aryptr();
5195   if (top_a  == NULL || top_a->klass()  == NULL ||
5196       top_b == NULL || top_b->klass()  == NULL ||
5197       top_n == NULL || top_n->klass()  == NULL ||
5198       top_m == NULL || top_m->klass()  == NULL) {
5199     // failed array check
5200     return false;
5201   }
5202 
5203   BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5204   BasicType b_elem = b_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5205   BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5206   BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5207   if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5208     return false;
5209   }
5210 
5211   // Make the call
5212   {
5213     Node* a_start = array_element_address(a, intcon(0), a_elem);
5214     Node* b_start = array_element_address(b, intcon(0), b_elem);
5215     Node* n_start = array_element_address(n, intcon(0), n_elem);
5216     Node* m_start = array_element_address(m, intcon(0), m_elem);
5217 
5218     Node* call = make_runtime_call(RC_LEAF,
5219                                    OptoRuntime::montgomeryMultiply_Type(),
5220                                    stubAddr, stubName, TypePtr::BOTTOM,
5221                                    a_start, b_start, n_start, len, inv, top(),
5222                                    m_start);
5223     set_result(m);
5224   }
5225 
5226   return true;
5227 }
5228 
5229 bool LibraryCallKit::inline_montgomerySquare() {
5230   address stubAddr = StubRoutines::montgomerySquare();
5231   if (stubAddr == NULL) {
5232     return false; // Intrinsic's stub is not implemented on this platform
5233   }
5234 
5235   assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
5236   const char* stubName = "montgomery_square";
5237 
5238   assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
5239 
5240   Node* a    = argument(0);
5241   Node* n    = argument(1);
5242   Node* len  = argument(2);
5243   Node* inv  = argument(3);
5244   Node* m    = argument(5);
5245 
5246   const Type* a_type = a->Value(&_gvn);
5247   const TypeAryPtr* top_a = a_type->isa_aryptr();
5248   const Type* n_type = a->Value(&_gvn);
5249   const TypeAryPtr* top_n = n_type->isa_aryptr();
5250   const Type* m_type = a->Value(&_gvn);
5251   const TypeAryPtr* top_m = m_type->isa_aryptr();
5252   if (top_a  == NULL || top_a->klass()  == NULL ||
5253       top_n == NULL || top_n->klass()  == NULL ||
5254       top_m == NULL || top_m->klass()  == NULL) {
5255     // failed array check
5256     return false;
5257   }
5258 
5259   BasicType a_elem = a_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5260   BasicType n_elem = n_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5261   BasicType m_elem = m_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5262   if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
5263     return false;
5264   }
5265 
5266   // Make the call
5267   {
5268     Node* a_start = array_element_address(a, intcon(0), a_elem);
5269     Node* n_start = array_element_address(n, intcon(0), n_elem);
5270     Node* m_start = array_element_address(m, intcon(0), m_elem);
5271 
5272     Node* call = make_runtime_call(RC_LEAF,
5273                                    OptoRuntime::montgomerySquare_Type(),
5274                                    stubAddr, stubName, TypePtr::BOTTOM,
5275                                    a_start, n_start, len, inv, top(),
5276                                    m_start);
5277     set_result(m);
5278   }
5279 
5280   return true;
5281 }
5282 
5283 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
5284   address stubAddr = NULL;
5285   const char* stubName = NULL;
5286 
5287   stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
5288   if (stubAddr == NULL) {
5289     return false; // Intrinsic's stub is not implemented on this platform
5290   }
5291 
5292   stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
5293 
5294   assert(callee()->signature()->size() == 5, "expected 5 arguments");
5295 
5296   Node* newArr = argument(0);
5297   Node* oldArr = argument(1);
5298   Node* newIdx = argument(2);
5299   Node* shiftCount = argument(3);
5300   Node* numIter = argument(4);
5301 
5302   const Type* newArr_type = newArr->Value(&_gvn);
5303   const TypeAryPtr* top_newArr = newArr_type->isa_aryptr();
5304   const Type* oldArr_type = oldArr->Value(&_gvn);
5305   const TypeAryPtr* top_oldArr = oldArr_type->isa_aryptr();
5306   if (top_newArr == NULL || top_newArr->klass() == NULL || top_oldArr == NULL
5307       || top_oldArr->klass() == NULL) {
5308     return false;
5309   }
5310 
5311   BasicType newArr_elem = newArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5312   BasicType oldArr_elem = oldArr_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5313   if (newArr_elem != T_INT || oldArr_elem != T_INT) {
5314     return false;
5315   }
5316 
5317   // Make the call
5318   {
5319     Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
5320     Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
5321 
5322     Node* call = make_runtime_call(RC_LEAF,
5323                                    OptoRuntime::bigIntegerShift_Type(),
5324                                    stubAddr,
5325                                    stubName,
5326                                    TypePtr::BOTTOM,
5327                                    newArr_start,
5328                                    oldArr_start,
5329                                    newIdx,
5330                                    shiftCount,
5331                                    numIter);
5332   }
5333 
5334   return true;
5335 }
5336 
5337 //-------------inline_vectorizedMismatch------------------------------
5338 bool LibraryCallKit::inline_vectorizedMismatch() {
5339   assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
5340 
5341   assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
5342   Node* obja    = argument(0); // Object
5343   Node* aoffset = argument(1); // long
5344   Node* objb    = argument(3); // Object
5345   Node* boffset = argument(4); // long
5346   Node* length  = argument(6); // int
5347   Node* scale   = argument(7); // int
5348 
5349   const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
5350   const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
5351   if (obja_t == NULL || obja_t->klass() == NULL ||
5352       objb_t == NULL || objb_t->klass() == NULL ||
5353       scale == top()) {
5354     return false; // failed input validation
5355   }
5356 
5357   Node* obja_adr = make_unsafe_address(obja, aoffset);
5358   Node* objb_adr = make_unsafe_address(objb, boffset);
5359 
5360   // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
5361   //
5362   //    inline_limit = ArrayOperationPartialInlineSize / element_size;
5363   //    if (length <= inline_limit) {
5364   //      inline_path:
5365   //        vmask   = VectorMaskGen length
5366   //        vload1  = LoadVectorMasked obja, vmask
5367   //        vload2  = LoadVectorMasked objb, vmask
5368   //        result1 = VectorCmpMasked vload1, vload2, vmask
5369   //    } else {
5370   //      call_stub_path:
5371   //        result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
5372   //    }
5373   //    exit_block:
5374   //      return Phi(result1, result2);
5375   //
5376   enum { inline_path = 1,  // input is small enough to process it all at once
5377          stub_path   = 2,  // input is too large; call into the VM
5378          PATH_LIMIT  = 3
5379   };
5380 
5381   Node* exit_block = new RegionNode(PATH_LIMIT);
5382   Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
5383   Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
5384 
5385   Node* call_stub_path = control();
5386 
5387   BasicType elem_bt = T_ILLEGAL;
5388 
5389   const TypeInt* scale_t = _gvn.type(scale)->is_int();
5390   if (scale_t->is_con()) {
5391     switch (scale_t->get_con()) {
5392       case 0: elem_bt = T_BYTE;  break;
5393       case 1: elem_bt = T_SHORT; break;
5394       case 2: elem_bt = T_INT;   break;
5395       case 3: elem_bt = T_LONG;  break;
5396 
5397       default: elem_bt = T_ILLEGAL; break; // not supported
5398     }
5399   }
5400 
5401   int inline_limit = 0;
5402   bool do_partial_inline = false;
5403 
5404   if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
5405     inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
5406     do_partial_inline = inline_limit >= 16;
5407   }
5408 
5409   if (do_partial_inline) {
5410     assert(elem_bt != T_ILLEGAL, "sanity");
5411 
5412     if (Matcher::match_rule_supported_vector(Op_VectorMaskGen,    inline_limit, elem_bt) &&
5413         Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
5414         Matcher::match_rule_supported_vector(Op_VectorCmpMasked,  inline_limit, elem_bt)) {
5415 
5416       const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
5417       Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
5418       Node* bol_gt     = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
5419 
5420       call_stub_path = generate_guard(bol_gt, NULL, PROB_MIN);
5421 
5422       if (!stopped()) {
5423         Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
5424 
5425         const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
5426         const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
5427         Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
5428         Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
5429 
5430         Node* vmask      = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
5431         Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
5432         Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
5433         Node* result     = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
5434 
5435         exit_block->init_req(inline_path, control());
5436         memory_phi->init_req(inline_path, map()->memory());
5437         result_phi->init_req(inline_path, result);
5438 
5439         C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
5440         clear_upper_avx();
5441       }
5442     }
5443   }
5444 
5445   if (call_stub_path != NULL) {
5446     set_control(call_stub_path);
5447 
5448     Node* call = make_runtime_call(RC_LEAF,
5449                                    OptoRuntime::vectorizedMismatch_Type(),
5450                                    StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
5451                                    obja_adr, objb_adr, length, scale);
5452 
5453     exit_block->init_req(stub_path, control());
5454     memory_phi->init_req(stub_path, map()->memory());
5455     result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
5456   }
5457 
5458   exit_block = _gvn.transform(exit_block);
5459   memory_phi = _gvn.transform(memory_phi);
5460   result_phi = _gvn.transform(result_phi);
5461 
5462   set_control(exit_block);
5463   set_all_memory(memory_phi);
5464   set_result(result_phi);
5465 
5466   return true;
5467 }
5468 
5469 /**
5470  * Calculate CRC32 for byte.
5471  * int java.util.zip.CRC32.update(int crc, int b)
5472  */
5473 bool LibraryCallKit::inline_updateCRC32() {
5474   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5475   assert(callee()->signature()->size() == 2, "update has 2 parameters");
5476   // no receiver since it is static method
5477   Node* crc  = argument(0); // type: int
5478   Node* b    = argument(1); // type: int
5479 
5480   /*
5481    *    int c = ~ crc;
5482    *    b = timesXtoThe32[(b ^ c) & 0xFF];
5483    *    b = b ^ (c >>> 8);
5484    *    crc = ~b;
5485    */
5486 
5487   Node* M1 = intcon(-1);
5488   crc = _gvn.transform(new XorINode(crc, M1));
5489   Node* result = _gvn.transform(new XorINode(crc, b));
5490   result = _gvn.transform(new AndINode(result, intcon(0xFF)));
5491 
5492   Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
5493   Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
5494   Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
5495   result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
5496 
5497   crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
5498   result = _gvn.transform(new XorINode(crc, result));
5499   result = _gvn.transform(new XorINode(result, M1));
5500   set_result(result);
5501   return true;
5502 }
5503 
5504 /**
5505  * Calculate CRC32 for byte[] array.
5506  * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
5507  */
5508 bool LibraryCallKit::inline_updateBytesCRC32() {
5509   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5510   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5511   // no receiver since it is static method
5512   Node* crc     = argument(0); // type: int
5513   Node* src     = argument(1); // type: oop
5514   Node* offset  = argument(2); // type: int
5515   Node* length  = argument(3); // type: int
5516 
5517   const Type* src_type = src->Value(&_gvn);
5518   const TypeAryPtr* top_src = src_type->isa_aryptr();
5519   if (top_src  == NULL || top_src->klass()  == NULL) {
5520     // failed array check
5521     return false;
5522   }
5523 
5524   // Figure out the size and type of the elements we will be copying.
5525   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5526   if (src_elem != T_BYTE) {
5527     return false;
5528   }
5529 
5530   // 'src_start' points to src array + scaled offset
5531   src = must_be_not_null(src, true);
5532   Node* src_start = array_element_address(src, offset, src_elem);
5533 
5534   // We assume that range check is done by caller.
5535   // TODO: generate range check (offset+length < src.length) in debug VM.
5536 
5537   // Call the stub.
5538   address stubAddr = StubRoutines::updateBytesCRC32();
5539   const char *stubName = "updateBytesCRC32";
5540 
5541   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5542                                  stubAddr, stubName, TypePtr::BOTTOM,
5543                                  crc, src_start, length);
5544   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5545   set_result(result);
5546   return true;
5547 }
5548 
5549 /**
5550  * Calculate CRC32 for ByteBuffer.
5551  * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
5552  */
5553 bool LibraryCallKit::inline_updateByteBufferCRC32() {
5554   assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
5555   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5556   // no receiver since it is static method
5557   Node* crc     = argument(0); // type: int
5558   Node* src     = argument(1); // type: long
5559   Node* offset  = argument(3); // type: int
5560   Node* length  = argument(4); // type: int
5561 
5562   src = ConvL2X(src);  // adjust Java long to machine word
5563   Node* base = _gvn.transform(new CastX2PNode(src));
5564   offset = ConvI2X(offset);
5565 
5566   // 'src_start' points to src array + scaled offset
5567   Node* src_start = basic_plus_adr(top(), base, offset);
5568 
5569   // Call the stub.
5570   address stubAddr = StubRoutines::updateBytesCRC32();
5571   const char *stubName = "updateBytesCRC32";
5572 
5573   Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
5574                                  stubAddr, stubName, TypePtr::BOTTOM,
5575                                  crc, src_start, length);
5576   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5577   set_result(result);
5578   return true;
5579 }
5580 
5581 //------------------------------get_table_from_crc32c_class-----------------------
5582 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
5583   Node* table = load_field_from_object(NULL, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
5584   assert (table != NULL, "wrong version of java.util.zip.CRC32C");
5585 
5586   return table;
5587 }
5588 
5589 //------------------------------inline_updateBytesCRC32C-----------------------
5590 //
5591 // Calculate CRC32C for byte[] array.
5592 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
5593 //
5594 bool LibraryCallKit::inline_updateBytesCRC32C() {
5595   assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5596   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5597   assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5598   // no receiver since it is a static method
5599   Node* crc     = argument(0); // type: int
5600   Node* src     = argument(1); // type: oop
5601   Node* offset  = argument(2); // type: int
5602   Node* end     = argument(3); // type: int
5603 
5604   Node* length = _gvn.transform(new SubINode(end, offset));
5605 
5606   const Type* src_type = src->Value(&_gvn);
5607   const TypeAryPtr* top_src = src_type->isa_aryptr();
5608   if (top_src  == NULL || top_src->klass()  == NULL) {
5609     // failed array check
5610     return false;
5611   }
5612 
5613   // Figure out the size and type of the elements we will be copying.
5614   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5615   if (src_elem != T_BYTE) {
5616     return false;
5617   }
5618 
5619   // 'src_start' points to src array + scaled offset
5620   src = must_be_not_null(src, true);
5621   Node* src_start = array_element_address(src, offset, src_elem);
5622 
5623   // static final int[] byteTable in class CRC32C
5624   Node* table = get_table_from_crc32c_class(callee()->holder());
5625   table = must_be_not_null(table, true);
5626   Node* table_start = array_element_address(table, intcon(0), T_INT);
5627 
5628   // We assume that range check is done by caller.
5629   // TODO: generate range check (offset+length < src.length) in debug VM.
5630 
5631   // Call the stub.
5632   address stubAddr = StubRoutines::updateBytesCRC32C();
5633   const char *stubName = "updateBytesCRC32C";
5634 
5635   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5636                                  stubAddr, stubName, TypePtr::BOTTOM,
5637                                  crc, src_start, length, table_start);
5638   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5639   set_result(result);
5640   return true;
5641 }
5642 
5643 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
5644 //
5645 // Calculate CRC32C for DirectByteBuffer.
5646 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
5647 //
5648 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
5649   assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
5650   assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
5651   assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
5652   // no receiver since it is a static method
5653   Node* crc     = argument(0); // type: int
5654   Node* src     = argument(1); // type: long
5655   Node* offset  = argument(3); // type: int
5656   Node* end     = argument(4); // type: int
5657 
5658   Node* length = _gvn.transform(new SubINode(end, offset));
5659 
5660   src = ConvL2X(src);  // adjust Java long to machine word
5661   Node* base = _gvn.transform(new CastX2PNode(src));
5662   offset = ConvI2X(offset);
5663 
5664   // 'src_start' points to src array + scaled offset
5665   Node* src_start = basic_plus_adr(top(), base, offset);
5666 
5667   // static final int[] byteTable in class CRC32C
5668   Node* table = get_table_from_crc32c_class(callee()->holder());
5669   table = must_be_not_null(table, true);
5670   Node* table_start = array_element_address(table, intcon(0), T_INT);
5671 
5672   // Call the stub.
5673   address stubAddr = StubRoutines::updateBytesCRC32C();
5674   const char *stubName = "updateBytesCRC32C";
5675 
5676   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
5677                                  stubAddr, stubName, TypePtr::BOTTOM,
5678                                  crc, src_start, length, table_start);
5679   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5680   set_result(result);
5681   return true;
5682 }
5683 
5684 //------------------------------inline_updateBytesAdler32----------------------
5685 //
5686 // Calculate Adler32 checksum for byte[] array.
5687 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
5688 //
5689 bool LibraryCallKit::inline_updateBytesAdler32() {
5690   assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5691   assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
5692   assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5693   // no receiver since it is static method
5694   Node* crc     = argument(0); // type: int
5695   Node* src     = argument(1); // type: oop
5696   Node* offset  = argument(2); // type: int
5697   Node* length  = argument(3); // type: int
5698 
5699   const Type* src_type = src->Value(&_gvn);
5700   const TypeAryPtr* top_src = src_type->isa_aryptr();
5701   if (top_src  == NULL || top_src->klass()  == NULL) {
5702     // failed array check
5703     return false;
5704   }
5705 
5706   // Figure out the size and type of the elements we will be copying.
5707   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
5708   if (src_elem != T_BYTE) {
5709     return false;
5710   }
5711 
5712   // 'src_start' points to src array + scaled offset
5713   Node* src_start = array_element_address(src, offset, src_elem);
5714 
5715   // We assume that range check is done by caller.
5716   // TODO: generate range check (offset+length < src.length) in debug VM.
5717 
5718   // Call the stub.
5719   address stubAddr = StubRoutines::updateBytesAdler32();
5720   const char *stubName = "updateBytesAdler32";
5721 
5722   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5723                                  stubAddr, stubName, TypePtr::BOTTOM,
5724                                  crc, src_start, length);
5725   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5726   set_result(result);
5727   return true;
5728 }
5729 
5730 //------------------------------inline_updateByteBufferAdler32---------------
5731 //
5732 // Calculate Adler32 checksum for DirectByteBuffer.
5733 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
5734 //
5735 bool LibraryCallKit::inline_updateByteBufferAdler32() {
5736   assert(UseAdler32Intrinsics, "Adler32 Instrinsic support need"); // check if we actually need to check this flag or check a different one
5737   assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
5738   assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
5739   // no receiver since it is static method
5740   Node* crc     = argument(0); // type: int
5741   Node* src     = argument(1); // type: long
5742   Node* offset  = argument(3); // type: int
5743   Node* length  = argument(4); // type: int
5744 
5745   src = ConvL2X(src);  // adjust Java long to machine word
5746   Node* base = _gvn.transform(new CastX2PNode(src));
5747   offset = ConvI2X(offset);
5748 
5749   // 'src_start' points to src array + scaled offset
5750   Node* src_start = basic_plus_adr(top(), base, offset);
5751 
5752   // Call the stub.
5753   address stubAddr = StubRoutines::updateBytesAdler32();
5754   const char *stubName = "updateBytesAdler32";
5755 
5756   Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
5757                                  stubAddr, stubName, TypePtr::BOTTOM,
5758                                  crc, src_start, length);
5759 
5760   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
5761   set_result(result);
5762   return true;
5763 }
5764 
5765 //----------------------------inline_reference_get----------------------------
5766 // public T java.lang.ref.Reference.get();
5767 bool LibraryCallKit::inline_reference_get() {
5768   const int referent_offset = java_lang_ref_Reference::referent_offset();
5769 
5770   // Get the argument:
5771   Node* reference_obj = null_check_receiver();
5772   if (stopped()) return true;
5773 
5774   DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
5775   Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
5776                                         decorators, /*is_static*/ false, NULL);
5777   if (result == NULL) return false;
5778 
5779   // Add memory barrier to prevent commoning reads from this field
5780   // across safepoint since GC can change its value.
5781   insert_mem_bar(Op_MemBarCPUOrder);
5782 
5783   set_result(result);
5784   return true;
5785 }
5786 
5787 //----------------------------inline_reference_refersTo0----------------------------
5788 // bool java.lang.ref.Reference.refersTo0();
5789 // bool java.lang.ref.PhantomReference.refersTo0();
5790 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
5791   // Get arguments:
5792   Node* reference_obj = null_check_receiver();
5793   Node* other_obj = argument(1);
5794   if (stopped()) return true;
5795 
5796   DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
5797   decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
5798   Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
5799                                           decorators, /*is_static*/ false, NULL);
5800   if (referent == NULL) return false;
5801 
5802   // Add memory barrier to prevent commoning reads from this field
5803   // across safepoint since GC can change its value.
5804   insert_mem_bar(Op_MemBarCPUOrder);
5805 
5806   Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
5807   Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
5808   IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
5809 
5810   RegionNode* region = new RegionNode(3);
5811   PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
5812 
5813   Node* if_true = _gvn.transform(new IfTrueNode(if_node));
5814   region->init_req(1, if_true);
5815   phi->init_req(1, intcon(1));
5816 
5817   Node* if_false = _gvn.transform(new IfFalseNode(if_node));
5818   region->init_req(2, if_false);
5819   phi->init_req(2, intcon(0));
5820 
5821   set_control(_gvn.transform(region));
5822   record_for_igvn(region);
5823   set_result(_gvn.transform(phi));
5824   return true;
5825 }
5826 
5827 
5828 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
5829                                              DecoratorSet decorators = IN_HEAP, bool is_static = false,
5830                                              ciInstanceKlass* fromKls = NULL) {
5831   if (fromKls == NULL) {
5832     const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5833     assert(tinst != NULL, "obj is null");
5834     assert(tinst->klass()->is_loaded(), "obj is not loaded");
5835     fromKls = tinst->klass()->as_instance_klass();
5836   } else {
5837     assert(is_static, "only for static field access");
5838   }
5839   ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5840                                               ciSymbol::make(fieldTypeString),
5841                                               is_static);
5842 
5843   assert (field != NULL, "undefined field");
5844   if (field == NULL) return (Node *) NULL;
5845 
5846   if (is_static) {
5847     const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5848     fromObj = makecon(tip);
5849   }
5850 
5851   // Next code  copied from Parse::do_get_xxx():
5852 
5853   // Compute address and memory type.
5854   int offset  = field->offset_in_bytes();
5855   bool is_vol = field->is_volatile();
5856   ciType* field_klass = field->type();
5857   assert(field_klass->is_loaded(), "should be loaded");
5858   const TypePtr* adr_type = C->alias_type(field)->adr_type();
5859   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5860   BasicType bt = field->layout_type();
5861 
5862   // Build the resultant type of the load
5863   const Type *type;
5864   if (bt == T_OBJECT) {
5865     type = TypeOopPtr::make_from_klass(field_klass->as_klass());
5866   } else {
5867     type = Type::get_const_basic_type(bt);
5868   }
5869 
5870   if (is_vol) {
5871     decorators |= MO_SEQ_CST;
5872   }
5873 
5874   return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
5875 }
5876 
5877 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
5878                                                  bool is_exact = true, bool is_static = false,
5879                                                  ciInstanceKlass * fromKls = NULL) {
5880   if (fromKls == NULL) {
5881     const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
5882     assert(tinst != NULL, "obj is null");
5883     assert(tinst->klass()->is_loaded(), "obj is not loaded");
5884     assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
5885     fromKls = tinst->klass()->as_instance_klass();
5886   }
5887   else {
5888     assert(is_static, "only for static field access");
5889   }
5890   ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
5891     ciSymbol::make(fieldTypeString),
5892     is_static);
5893 
5894   assert(field != NULL, "undefined field");
5895   assert(!field->is_volatile(), "not defined for volatile fields");
5896 
5897   if (is_static) {
5898     const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
5899     fromObj = makecon(tip);
5900   }
5901 
5902   // Next code  copied from Parse::do_get_xxx():
5903 
5904   // Compute address and memory type.
5905   int offset = field->offset_in_bytes();
5906   Node *adr = basic_plus_adr(fromObj, fromObj, offset);
5907 
5908   return adr;
5909 }
5910 
5911 //------------------------------inline_aescrypt_Block-----------------------
5912 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
5913   address stubAddr = NULL;
5914   const char *stubName;
5915   assert(UseAES, "need AES instruction support");
5916 
5917   switch(id) {
5918   case vmIntrinsics::_aescrypt_encryptBlock:
5919     stubAddr = StubRoutines::aescrypt_encryptBlock();
5920     stubName = "aescrypt_encryptBlock";
5921     break;
5922   case vmIntrinsics::_aescrypt_decryptBlock:
5923     stubAddr = StubRoutines::aescrypt_decryptBlock();
5924     stubName = "aescrypt_decryptBlock";
5925     break;
5926   default:
5927     break;
5928   }
5929   if (stubAddr == NULL) return false;
5930 
5931   Node* aescrypt_object = argument(0);
5932   Node* src             = argument(1);
5933   Node* src_offset      = argument(2);
5934   Node* dest            = argument(3);
5935   Node* dest_offset     = argument(4);
5936 
5937   src = must_be_not_null(src, true);
5938   dest = must_be_not_null(dest, true);
5939 
5940   // (1) src and dest are arrays.
5941   const Type* src_type = src->Value(&_gvn);
5942   const Type* dest_type = dest->Value(&_gvn);
5943   const TypeAryPtr* top_src = src_type->isa_aryptr();
5944   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5945   assert (top_src  != NULL && top_src->klass()  != NULL &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
5946 
5947   // for the quick and dirty code we will skip all the checks.
5948   // we are just trying to get the call to be generated.
5949   Node* src_start  = src;
5950   Node* dest_start = dest;
5951   if (src_offset != NULL || dest_offset != NULL) {
5952     assert(src_offset != NULL && dest_offset != NULL, "");
5953     src_start  = array_element_address(src,  src_offset,  T_BYTE);
5954     dest_start = array_element_address(dest, dest_offset, T_BYTE);
5955   }
5956 
5957   // now need to get the start of its expanded key array
5958   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
5959   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
5960   if (k_start == NULL) return false;
5961 
5962   // Call the stub.
5963   make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
5964                     stubAddr, stubName, TypePtr::BOTTOM,
5965                     src_start, dest_start, k_start);
5966 
5967   return true;
5968 }
5969 
5970 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
5971 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
5972   address stubAddr = NULL;
5973   const char *stubName = NULL;
5974 
5975   assert(UseAES, "need AES instruction support");
5976 
5977   switch(id) {
5978   case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
5979     stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
5980     stubName = "cipherBlockChaining_encryptAESCrypt";
5981     break;
5982   case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
5983     stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
5984     stubName = "cipherBlockChaining_decryptAESCrypt";
5985     break;
5986   default:
5987     break;
5988   }
5989   if (stubAddr == NULL) return false;
5990 
5991   Node* cipherBlockChaining_object = argument(0);
5992   Node* src                        = argument(1);
5993   Node* src_offset                 = argument(2);
5994   Node* len                        = argument(3);
5995   Node* dest                       = argument(4);
5996   Node* dest_offset                = argument(5);
5997 
5998   src = must_be_not_null(src, false);
5999   dest = must_be_not_null(dest, false);
6000 
6001   // (1) src and dest are arrays.
6002   const Type* src_type = src->Value(&_gvn);
6003   const Type* dest_type = dest->Value(&_gvn);
6004   const TypeAryPtr* top_src = src_type->isa_aryptr();
6005   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6006   assert (top_src  != NULL && top_src->klass()  != NULL
6007           &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6008 
6009   // checks are the responsibility of the caller
6010   Node* src_start  = src;
6011   Node* dest_start = dest;
6012   if (src_offset != NULL || dest_offset != NULL) {
6013     assert(src_offset != NULL && dest_offset != NULL, "");
6014     src_start  = array_element_address(src,  src_offset,  T_BYTE);
6015     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6016   }
6017 
6018   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6019   // (because of the predicated logic executed earlier).
6020   // so we cast it here safely.
6021   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6022 
6023   Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6024   if (embeddedCipherObj == NULL) return false;
6025 
6026   // cast it to what we know it will be at runtime
6027   const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6028   assert(tinst != NULL, "CBC obj is null");
6029   assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
6030   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6031   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6032 
6033   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6034   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6035   const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6036   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6037   aescrypt_object = _gvn.transform(aescrypt_object);
6038 
6039   // we need to get the start of the aescrypt_object's expanded key array
6040   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6041   if (k_start == NULL) return false;
6042 
6043   // similarly, get the start address of the r vector
6044   Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6045   if (objRvec == NULL) return false;
6046   Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6047 
6048   // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6049   Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6050                                      OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6051                                      stubAddr, stubName, TypePtr::BOTTOM,
6052                                      src_start, dest_start, k_start, r_start, len);
6053 
6054   // return cipher length (int)
6055   Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
6056   set_result(retvalue);
6057   return true;
6058 }
6059 
6060 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
6061 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
6062   address stubAddr = NULL;
6063   const char *stubName = NULL;
6064 
6065   assert(UseAES, "need AES instruction support");
6066 
6067   switch (id) {
6068   case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
6069     stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
6070     stubName = "electronicCodeBook_encryptAESCrypt";
6071     break;
6072   case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
6073     stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
6074     stubName = "electronicCodeBook_decryptAESCrypt";
6075     break;
6076   default:
6077     break;
6078   }
6079 
6080   if (stubAddr == NULL) return false;
6081 
6082   Node* electronicCodeBook_object = argument(0);
6083   Node* src                       = argument(1);
6084   Node* src_offset                = argument(2);
6085   Node* len                       = argument(3);
6086   Node* dest                      = argument(4);
6087   Node* dest_offset               = argument(5);
6088 
6089   // (1) src and dest are arrays.
6090   const Type* src_type = src->Value(&_gvn);
6091   const Type* dest_type = dest->Value(&_gvn);
6092   const TypeAryPtr* top_src = src_type->isa_aryptr();
6093   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6094   assert(top_src != NULL && top_src->klass() != NULL
6095          &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6096 
6097   // checks are the responsibility of the caller
6098   Node* src_start = src;
6099   Node* dest_start = dest;
6100   if (src_offset != NULL || dest_offset != NULL) {
6101     assert(src_offset != NULL && dest_offset != NULL, "");
6102     src_start = array_element_address(src, src_offset, T_BYTE);
6103     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6104   }
6105 
6106   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6107   // (because of the predicated logic executed earlier).
6108   // so we cast it here safely.
6109   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6110 
6111   Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6112   if (embeddedCipherObj == NULL) return false;
6113 
6114   // cast it to what we know it will be at runtime
6115   const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
6116   assert(tinst != NULL, "ECB obj is null");
6117   assert(tinst->klass()->is_loaded(), "ECB obj is not loaded");
6118   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6119   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6120 
6121   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6122   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6123   const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6124   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6125   aescrypt_object = _gvn.transform(aescrypt_object);
6126 
6127   // we need to get the start of the aescrypt_object's expanded key array
6128   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6129   if (k_start == NULL) return false;
6130 
6131   // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6132   Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
6133                                      OptoRuntime::electronicCodeBook_aescrypt_Type(),
6134                                      stubAddr, stubName, TypePtr::BOTTOM,
6135                                      src_start, dest_start, k_start, len);
6136 
6137   // return cipher length (int)
6138   Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
6139   set_result(retvalue);
6140   return true;
6141 }
6142 
6143 //------------------------------inline_counterMode_AESCrypt-----------------------
6144 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
6145   assert(UseAES, "need AES instruction support");
6146   if (!UseAESCTRIntrinsics) return false;
6147 
6148   address stubAddr = NULL;
6149   const char *stubName = NULL;
6150   if (id == vmIntrinsics::_counterMode_AESCrypt) {
6151     stubAddr = StubRoutines::counterMode_AESCrypt();
6152     stubName = "counterMode_AESCrypt";
6153   }
6154   if (stubAddr == NULL) return false;
6155 
6156   Node* counterMode_object = argument(0);
6157   Node* src = argument(1);
6158   Node* src_offset = argument(2);
6159   Node* len = argument(3);
6160   Node* dest = argument(4);
6161   Node* dest_offset = argument(5);
6162 
6163   // (1) src and dest are arrays.
6164   const Type* src_type = src->Value(&_gvn);
6165   const Type* dest_type = dest->Value(&_gvn);
6166   const TypeAryPtr* top_src = src_type->isa_aryptr();
6167   const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6168   assert(top_src != NULL && top_src->klass() != NULL &&
6169          top_dest != NULL && top_dest->klass() != NULL, "args are strange");
6170 
6171   // checks are the responsibility of the caller
6172   Node* src_start = src;
6173   Node* dest_start = dest;
6174   if (src_offset != NULL || dest_offset != NULL) {
6175     assert(src_offset != NULL && dest_offset != NULL, "");
6176     src_start = array_element_address(src, src_offset, T_BYTE);
6177     dest_start = array_element_address(dest, dest_offset, T_BYTE);
6178   }
6179 
6180   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6181   // (because of the predicated logic executed earlier).
6182   // so we cast it here safely.
6183   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6184   Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6185   if (embeddedCipherObj == NULL) return false;
6186   // cast it to what we know it will be at runtime
6187   const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
6188   assert(tinst != NULL, "CTR obj is null");
6189   assert(tinst->klass()->is_loaded(), "CTR obj is not loaded");
6190   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6191   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6192   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6193   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6194   const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6195   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6196   aescrypt_object = _gvn.transform(aescrypt_object);
6197   // we need to get the start of the aescrypt_object's expanded key array
6198   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6199   if (k_start == NULL) return false;
6200   // similarly, get the start address of the r vector
6201   Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
6202   if (obj_counter == NULL) return false;
6203   Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
6204 
6205   Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
6206   if (saved_encCounter == NULL) return false;
6207   Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
6208   Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
6209 
6210   // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6211   Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6212                                      OptoRuntime::counterMode_aescrypt_Type(),
6213                                      stubAddr, stubName, TypePtr::BOTTOM,
6214                                      src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
6215 
6216   // return cipher length (int)
6217   Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
6218   set_result(retvalue);
6219   return true;
6220 }
6221 
6222 //------------------------------get_key_start_from_aescrypt_object-----------------------
6223 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
6224 #if defined(PPC64) || defined(S390)
6225   // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
6226   // Intel's extention is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
6227   // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
6228   // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
6229   Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
6230   assert (objSessionK != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6231   if (objSessionK == NULL) {
6232     return (Node *) NULL;
6233   }
6234   Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
6235 #else
6236   Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
6237 #endif // PPC64
6238   assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
6239   if (objAESCryptKey == NULL) return (Node *) NULL;
6240 
6241   // now have the array, need to get the start address of the K array
6242   Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
6243   return k_start;
6244 }
6245 
6246 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
6247 // Return node representing slow path of predicate check.
6248 // the pseudo code we want to emulate with this predicate is:
6249 // for encryption:
6250 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6251 // for decryption:
6252 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6253 //    note cipher==plain is more conservative than the original java code but that's OK
6254 //
6255 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
6256   // The receiver was checked for NULL already.
6257   Node* objCBC = argument(0);
6258 
6259   Node* src = argument(1);
6260   Node* dest = argument(4);
6261 
6262   // Load embeddedCipher field of CipherBlockChaining object.
6263   Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6264 
6265   // get AESCrypt klass for instanceOf check
6266   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6267   // will have same classloader as CipherBlockChaining object
6268   const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
6269   assert(tinst != NULL, "CBCobj is null");
6270   assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");
6271 
6272   // we want to do an instanceof comparison against the AESCrypt class
6273   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6274   if (!klass_AESCrypt->is_loaded()) {
6275     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6276     Node* ctrl = control();
6277     set_control(top()); // no regular fast path
6278     return ctrl;
6279   }
6280 
6281   src = must_be_not_null(src, true);
6282   dest = must_be_not_null(dest, true);
6283 
6284   // Resolve oops to stable for CmpP below.
6285   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6286 
6287   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6288   Node* cmp_instof  = _gvn.transform(new CmpINode(instof, intcon(1)));
6289   Node* bool_instof  = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6290 
6291   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6292 
6293   // for encryption, we are done
6294   if (!decrypting)
6295     return instof_false;  // even if it is NULL
6296 
6297   // for decryption, we need to add a further check to avoid
6298   // taking the intrinsic path when cipher and plain are the same
6299   // see the original java code for why.
6300   RegionNode* region = new RegionNode(3);
6301   region->init_req(1, instof_false);
6302 
6303   Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6304   Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6305   Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6306   region->init_req(2, src_dest_conjoint);
6307 
6308   record_for_igvn(region);
6309   return _gvn.transform(region);
6310 }
6311 
6312 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
6313 // Return node representing slow path of predicate check.
6314 // the pseudo code we want to emulate with this predicate is:
6315 // for encryption:
6316 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6317 // for decryption:
6318 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6319 //    note cipher==plain is more conservative than the original java code but that's OK
6320 //
6321 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
6322   // The receiver was checked for NULL already.
6323   Node* objECB = argument(0);
6324 
6325   // Load embeddedCipher field of ElectronicCodeBook object.
6326   Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6327 
6328   // get AESCrypt klass for instanceOf check
6329   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6330   // will have same classloader as ElectronicCodeBook object
6331   const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
6332   assert(tinst != NULL, "ECBobj is null");
6333   assert(tinst->klass()->is_loaded(), "ECBobj is not loaded");
6334 
6335   // we want to do an instanceof comparison against the AESCrypt class
6336   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6337   if (!klass_AESCrypt->is_loaded()) {
6338     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6339     Node* ctrl = control();
6340     set_control(top()); // no regular fast path
6341     return ctrl;
6342   }
6343   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6344 
6345   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6346   Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6347   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6348 
6349   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6350 
6351   // for encryption, we are done
6352   if (!decrypting)
6353     return instof_false;  // even if it is NULL
6354 
6355   // for decryption, we need to add a further check to avoid
6356   // taking the intrinsic path when cipher and plain are the same
6357   // see the original java code for why.
6358   RegionNode* region = new RegionNode(3);
6359   region->init_req(1, instof_false);
6360   Node* src = argument(1);
6361   Node* dest = argument(4);
6362   Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
6363   Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
6364   Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
6365   region->init_req(2, src_dest_conjoint);
6366 
6367   record_for_igvn(region);
6368   return _gvn.transform(region);
6369 }
6370 
6371 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
6372 // Return node representing slow path of predicate check.
6373 // the pseudo code we want to emulate with this predicate is:
6374 // for encryption:
6375 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6376 // for decryption:
6377 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6378 //    note cipher==plain is more conservative than the original java code but that's OK
6379 //
6380 
6381 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
6382   // The receiver was checked for NULL already.
6383   Node* objCTR = argument(0);
6384 
6385   // Load embeddedCipher field of CipherBlockChaining object.
6386   Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6387 
6388   // get AESCrypt klass for instanceOf check
6389   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6390   // will have same classloader as CipherBlockChaining object
6391   const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
6392   assert(tinst != NULL, "CTRobj is null");
6393   assert(tinst->klass()->is_loaded(), "CTRobj is not loaded");
6394 
6395   // we want to do an instanceof comparison against the AESCrypt class
6396   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6397   if (!klass_AESCrypt->is_loaded()) {
6398     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6399     Node* ctrl = control();
6400     set_control(top()); // no regular fast path
6401     return ctrl;
6402   }
6403 
6404   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6405   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6406   Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6407   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6408   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6409 
6410   return instof_false; // even if it is NULL
6411 }
6412 
6413 //------------------------------inline_ghash_processBlocks
6414 bool LibraryCallKit::inline_ghash_processBlocks() {
6415   address stubAddr;
6416   const char *stubName;
6417   assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
6418 
6419   stubAddr = StubRoutines::ghash_processBlocks();
6420   stubName = "ghash_processBlocks";
6421 
6422   Node* data           = argument(0);
6423   Node* offset         = argument(1);
6424   Node* len            = argument(2);
6425   Node* state          = argument(3);
6426   Node* subkeyH        = argument(4);
6427 
6428   state = must_be_not_null(state, true);
6429   subkeyH = must_be_not_null(subkeyH, true);
6430   data = must_be_not_null(data, true);
6431 
6432   Node* state_start  = array_element_address(state, intcon(0), T_LONG);
6433   assert(state_start, "state is NULL");
6434   Node* subkeyH_start  = array_element_address(subkeyH, intcon(0), T_LONG);
6435   assert(subkeyH_start, "subkeyH is NULL");
6436   Node* data_start  = array_element_address(data, offset, T_BYTE);
6437   assert(data_start, "data is NULL");
6438 
6439   Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
6440                                   OptoRuntime::ghash_processBlocks_Type(),
6441                                   stubAddr, stubName, TypePtr::BOTTOM,
6442                                   state_start, subkeyH_start, data_start, len);
6443   return true;
6444 }
6445 
6446 bool LibraryCallKit::inline_base64_encodeBlock() {
6447   address stubAddr;
6448   const char *stubName;
6449   assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6450   assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
6451   stubAddr = StubRoutines::base64_encodeBlock();
6452   stubName = "encodeBlock";
6453 
6454   if (!stubAddr) return false;
6455   Node* base64obj = argument(0);
6456   Node* src = argument(1);
6457   Node* offset = argument(2);
6458   Node* len = argument(3);
6459   Node* dest = argument(4);
6460   Node* dp = argument(5);
6461   Node* isURL = argument(6);
6462 
6463   src = must_be_not_null(src, true);
6464   dest = must_be_not_null(dest, true);
6465 
6466   Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6467   assert(src_start, "source array is NULL");
6468   Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6469   assert(dest_start, "destination array is NULL");
6470 
6471   Node* base64 = make_runtime_call(RC_LEAF,
6472                                    OptoRuntime::base64_encodeBlock_Type(),
6473                                    stubAddr, stubName, TypePtr::BOTTOM,
6474                                    src_start, offset, len, dest_start, dp, isURL);
6475   return true;
6476 }
6477 
6478 bool LibraryCallKit::inline_base64_decodeBlock() {
6479   address stubAddr;
6480   const char *stubName;
6481   assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
6482   assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
6483   stubAddr = StubRoutines::base64_decodeBlock();
6484   stubName = "decodeBlock";
6485 
6486   if (!stubAddr) return false;
6487   Node* base64obj = argument(0);
6488   Node* src = argument(1);
6489   Node* src_offset = argument(2);
6490   Node* len = argument(3);
6491   Node* dest = argument(4);
6492   Node* dest_offset = argument(5);
6493   Node* isURL = argument(6);
6494   Node* isMIME = argument(7);
6495 
6496   src = must_be_not_null(src, true);
6497   dest = must_be_not_null(dest, true);
6498 
6499   Node* src_start = array_element_address(src, intcon(0), T_BYTE);
6500   assert(src_start, "source array is NULL");
6501   Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
6502   assert(dest_start, "destination array is NULL");
6503 
6504   Node* call = make_runtime_call(RC_LEAF,
6505                                  OptoRuntime::base64_decodeBlock_Type(),
6506                                  stubAddr, stubName, TypePtr::BOTTOM,
6507                                  src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
6508   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6509   set_result(result);
6510   return true;
6511 }
6512 
6513 //------------------------------inline_digestBase_implCompress-----------------------
6514 //
6515 // Calculate MD5 for single-block byte[] array.
6516 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
6517 //
6518 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
6519 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
6520 //
6521 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
6522 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
6523 //
6524 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
6525 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
6526 //
6527 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
6528 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
6529 //
6530 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
6531   assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
6532 
6533   Node* digestBase_obj = argument(0);
6534   Node* src            = argument(1); // type oop
6535   Node* ofs            = argument(2); // type int
6536 
6537   const Type* src_type = src->Value(&_gvn);
6538   const TypeAryPtr* top_src = src_type->isa_aryptr();
6539   if (top_src  == NULL || top_src->klass()  == NULL) {
6540     // failed array check
6541     return false;
6542   }
6543   // Figure out the size and type of the elements we will be copying.
6544   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6545   if (src_elem != T_BYTE) {
6546     return false;
6547   }
6548   // 'src_start' points to src array + offset
6549   src = must_be_not_null(src, true);
6550   Node* src_start = array_element_address(src, ofs, src_elem);
6551   Node* state = NULL;
6552   Node* digest_length = NULL;
6553   address stubAddr;
6554   const char *stubName;
6555 
6556   switch(id) {
6557   case vmIntrinsics::_md5_implCompress:
6558     assert(UseMD5Intrinsics, "need MD5 instruction support");
6559     state = get_state_from_digest_object(digestBase_obj, "[I");
6560     stubAddr = StubRoutines::md5_implCompress();
6561     stubName = "md5_implCompress";
6562     break;
6563   case vmIntrinsics::_sha_implCompress:
6564     assert(UseSHA1Intrinsics, "need SHA1 instruction support");
6565     state = get_state_from_digest_object(digestBase_obj, "[I");
6566     stubAddr = StubRoutines::sha1_implCompress();
6567     stubName = "sha1_implCompress";
6568     break;
6569   case vmIntrinsics::_sha2_implCompress:
6570     assert(UseSHA256Intrinsics, "need SHA256 instruction support");
6571     state = get_state_from_digest_object(digestBase_obj, "[I");
6572     stubAddr = StubRoutines::sha256_implCompress();
6573     stubName = "sha256_implCompress";
6574     break;
6575   case vmIntrinsics::_sha5_implCompress:
6576     assert(UseSHA512Intrinsics, "need SHA512 instruction support");
6577     state = get_state_from_digest_object(digestBase_obj, "[J");
6578     stubAddr = StubRoutines::sha512_implCompress();
6579     stubName = "sha512_implCompress";
6580     break;
6581   case vmIntrinsics::_sha3_implCompress:
6582     assert(UseSHA3Intrinsics, "need SHA3 instruction support");
6583     state = get_state_from_digest_object(digestBase_obj, "[B");
6584     stubAddr = StubRoutines::sha3_implCompress();
6585     stubName = "sha3_implCompress";
6586     digest_length = get_digest_length_from_digest_object(digestBase_obj);
6587     if (digest_length == NULL) return false;
6588     break;
6589   default:
6590     fatal_unexpected_iid(id);
6591     return false;
6592   }
6593   if (state == NULL) return false;
6594 
6595   assert(stubAddr != NULL, "Stub is generated");
6596   if (stubAddr == NULL) return false;
6597 
6598   // Call the stub.
6599   Node* call;
6600   if (digest_length == NULL) {
6601     call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
6602                              stubAddr, stubName, TypePtr::BOTTOM,
6603                              src_start, state);
6604   } else {
6605     call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
6606                              stubAddr, stubName, TypePtr::BOTTOM,
6607                              src_start, state, digest_length);
6608   }
6609 
6610   return true;
6611 }
6612 
6613 //------------------------------inline_digestBase_implCompressMB-----------------------
6614 //
6615 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
6616 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
6617 //
6618 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
6619   assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6620          "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6621   assert((uint)predicate < 5, "sanity");
6622   assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
6623 
6624   Node* digestBase_obj = argument(0); // The receiver was checked for NULL already.
6625   Node* src            = argument(1); // byte[] array
6626   Node* ofs            = argument(2); // type int
6627   Node* limit          = argument(3); // type int
6628 
6629   const Type* src_type = src->Value(&_gvn);
6630   const TypeAryPtr* top_src = src_type->isa_aryptr();
6631   if (top_src  == NULL || top_src->klass()  == NULL) {
6632     // failed array check
6633     return false;
6634   }
6635   // Figure out the size and type of the elements we will be copying.
6636   BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
6637   if (src_elem != T_BYTE) {
6638     return false;
6639   }
6640   // 'src_start' points to src array + offset
6641   src = must_be_not_null(src, false);
6642   Node* src_start = array_element_address(src, ofs, src_elem);
6643 
6644   const char* klass_digestBase_name = NULL;
6645   const char* stub_name = NULL;
6646   address     stub_addr = NULL;
6647   const char* state_type = "[I";
6648 
6649   switch (predicate) {
6650   case 0:
6651     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
6652       klass_digestBase_name = "sun/security/provider/MD5";
6653       stub_name = "md5_implCompressMB";
6654       stub_addr = StubRoutines::md5_implCompressMB();
6655     }
6656     break;
6657   case 1:
6658     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
6659       klass_digestBase_name = "sun/security/provider/SHA";
6660       stub_name = "sha1_implCompressMB";
6661       stub_addr = StubRoutines::sha1_implCompressMB();
6662     }
6663     break;
6664   case 2:
6665     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
6666       klass_digestBase_name = "sun/security/provider/SHA2";
6667       stub_name = "sha256_implCompressMB";
6668       stub_addr = StubRoutines::sha256_implCompressMB();
6669     }
6670     break;
6671   case 3:
6672     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
6673       klass_digestBase_name = "sun/security/provider/SHA5";
6674       stub_name = "sha512_implCompressMB";
6675       stub_addr = StubRoutines::sha512_implCompressMB();
6676       state_type = "[J";
6677     }
6678     break;
6679   case 4:
6680     if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
6681       klass_digestBase_name = "sun/security/provider/SHA3";
6682       stub_name = "sha3_implCompressMB";
6683       stub_addr = StubRoutines::sha3_implCompressMB();
6684       state_type = "[B";
6685     }
6686     break;
6687   default:
6688     fatal("unknown DigestBase intrinsic predicate: %d", predicate);
6689   }
6690   if (klass_digestBase_name != NULL) {
6691     assert(stub_addr != NULL, "Stub is generated");
6692     if (stub_addr == NULL) return false;
6693 
6694     // get DigestBase klass to lookup for SHA klass
6695     const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
6696     assert(tinst != NULL, "digestBase_obj is not instance???");
6697     assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6698 
6699     ciKlass* klass_digestBase = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
6700     assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
6701     ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
6702     return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, state_type, stub_addr, stub_name, src_start, ofs, limit);
6703   }
6704   return false;
6705 }
6706 
6707 //------------------------------inline_digestBase_implCompressMB-----------------------
6708 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
6709                                                       const char* state_type, address stubAddr, const char *stubName,
6710                                                       Node* src_start, Node* ofs, Node* limit) {
6711   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
6712   const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6713   Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
6714   digest_obj = _gvn.transform(digest_obj);
6715 
6716   Node* state = get_state_from_digest_object(digest_obj, state_type);
6717   if (state == NULL) return false;
6718 
6719   Node* digest_length = NULL;
6720   if (strcmp("sha3_implCompressMB", stubName) == 0) {
6721     digest_length = get_digest_length_from_digest_object(digest_obj);
6722     if (digest_length == NULL) return false;
6723   }
6724 
6725   // Call the stub.
6726   Node* call;
6727   if (digest_length == NULL) {
6728     call = make_runtime_call(RC_LEAF|RC_NO_FP,
6729                              OptoRuntime::digestBase_implCompressMB_Type(false),
6730                              stubAddr, stubName, TypePtr::BOTTOM,
6731                              src_start, state, ofs, limit);
6732   } else {
6733      call = make_runtime_call(RC_LEAF|RC_NO_FP,
6734                              OptoRuntime::digestBase_implCompressMB_Type(true),
6735                              stubAddr, stubName, TypePtr::BOTTOM,
6736                              src_start, state, digest_length, ofs, limit);
6737   }
6738 
6739   // return ofs (int)
6740   Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6741   set_result(result);
6742 
6743   return true;
6744 }
6745 
6746 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
6747 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
6748   assert(UseAES, "need AES instruction support");
6749   address stubAddr = NULL;
6750   const char *stubName = NULL;
6751   stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
6752   stubName = "galoisCounterMode_AESCrypt";
6753 
6754   if (stubAddr == NULL) return false;
6755 
6756   Node* in      = argument(0);
6757   Node* inOfs   = argument(1);
6758   Node* len     = argument(2);
6759   Node* ct      = argument(3);
6760   Node* ctOfs   = argument(4);
6761   Node* out     = argument(5);
6762   Node* outOfs  = argument(6);
6763   Node* gctr_object = argument(7);
6764   Node* ghash_object = argument(8);
6765 
6766   // (1) in, ct and out are arrays.
6767   const Type* in_type = in->Value(&_gvn);
6768   const Type* ct_type = ct->Value(&_gvn);
6769   const Type* out_type = out->Value(&_gvn);
6770   const TypeAryPtr* top_in = in_type->isa_aryptr();
6771   const TypeAryPtr* top_ct = ct_type->isa_aryptr();
6772   const TypeAryPtr* top_out = out_type->isa_aryptr();
6773   assert(top_in != NULL && top_in->klass() != NULL &&
6774          top_ct != NULL && top_ct->klass() != NULL &&
6775          top_out != NULL && top_out->klass() != NULL, "args are strange");
6776 
6777   // checks are the responsibility of the caller
6778   Node* in_start = in;
6779   Node* ct_start = ct;
6780   Node* out_start = out;
6781   if (inOfs != NULL || ctOfs != NULL || outOfs != NULL) {
6782     assert(inOfs != NULL && ctOfs != NULL && outOfs != NULL, "");
6783     in_start = array_element_address(in, inOfs, T_BYTE);
6784     ct_start = array_element_address(ct, ctOfs, T_BYTE);
6785     out_start = array_element_address(out, outOfs, T_BYTE);
6786   }
6787 
6788   // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6789   // (because of the predicated logic executed earlier).
6790   // so we cast it here safely.
6791   // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6792   Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6793   Node* counter = load_field_from_object(gctr_object, "counter", "[B");
6794   Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
6795   Node* state = load_field_from_object(ghash_object, "state", "[J");
6796 
6797   if (embeddedCipherObj == NULL || counter == NULL || subkeyHtbl == NULL || state == NULL) {
6798     return false;
6799   }
6800   // cast it to what we know it will be at runtime
6801   const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
6802   assert(tinst != NULL, "GCTR obj is null");
6803   assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
6804   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6805   assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6806   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6807   const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6808   const TypeOopPtr* xtype = aklass->as_instance_type();
6809   Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6810   aescrypt_object = _gvn.transform(aescrypt_object);
6811   // we need to get the start of the aescrypt_object's expanded key array
6812   Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6813   if (k_start == NULL) return false;
6814   // similarly, get the start address of the r vector
6815   Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
6816   Node* state_start = array_element_address(state, intcon(0), T_LONG);
6817   Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
6818 
6819 
6820   // Call the stub, passing params
6821   Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6822                                OptoRuntime::galoisCounterMode_aescrypt_Type(),
6823                                stubAddr, stubName, TypePtr::BOTTOM,
6824                                in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
6825 
6826   // return cipher length (int)
6827   Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
6828   set_result(retvalue);
6829 
6830   return true;
6831 }
6832 
6833 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
6834 // Return node representing slow path of predicate check.
6835 // the pseudo code we want to emulate with this predicate is:
6836 // for encryption:
6837 //    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
6838 // for decryption:
6839 //    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
6840 //    note cipher==plain is more conservative than the original java code but that's OK
6841 //
6842 
6843 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
6844   // The receiver was checked for NULL already.
6845   Node* objGCTR = argument(7);
6846   // Load embeddedCipher field of GCTR object.
6847   Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6848   assert(embeddedCipherObj != NULL, "embeddedCipherObj is null");
6849 
6850   // get AESCrypt klass for instanceOf check
6851   // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
6852   // will have same classloader as CipherBlockChaining object
6853   const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
6854   assert(tinst != NULL, "GCTR obj is null");
6855   assert(tinst->klass()->is_loaded(), "GCTR obj is not loaded");
6856 
6857   // we want to do an instanceof comparison against the AESCrypt class
6858   ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6859   if (!klass_AESCrypt->is_loaded()) {
6860     // if AESCrypt is not even loaded, we never take the intrinsic fast path
6861     Node* ctrl = control();
6862     set_control(top()); // no regular fast path
6863     return ctrl;
6864   }
6865 
6866   ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6867   Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
6868   Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6869   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6870   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6871 
6872   return instof_false; // even if it is NULL
6873 }
6874 
6875 //------------------------------get_state_from_digest_object-----------------------
6876 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, const char *state_type) {
6877   Node* digest_state = load_field_from_object(digest_object, "state", state_type);
6878   assert (digest_state != NULL, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
6879   if (digest_state == NULL) return (Node *) NULL;
6880 
6881   // now have the array, need to get the start address of the state array
6882   Node* state = array_element_address(digest_state, intcon(0), T_INT);
6883   return state;
6884 }
6885 
6886 //------------------------------get_digest_length_from_sha3_object----------------------------------
6887 Node * LibraryCallKit::get_digest_length_from_digest_object(Node *digest_object) {
6888   Node* digest_length = load_field_from_object(digest_object, "digestLength", "I");
6889   assert (digest_length != NULL, "sanity");
6890   return digest_length;
6891 }
6892 
6893 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
6894 // Return node representing slow path of predicate check.
6895 // the pseudo code we want to emulate with this predicate is:
6896 //    if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
6897 //
6898 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
6899   assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
6900          "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
6901   assert((uint)predicate < 5, "sanity");
6902 
6903   // The receiver was checked for NULL already.
6904   Node* digestBaseObj = argument(0);
6905 
6906   // get DigestBase klass for instanceOf check
6907   const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
6908   assert(tinst != NULL, "digestBaseObj is null");
6909   assert(tinst->klass()->is_loaded(), "DigestBase is not loaded");
6910 
6911   const char* klass_name = NULL;
6912   switch (predicate) {
6913   case 0:
6914     if (UseMD5Intrinsics) {
6915       // we want to do an instanceof comparison against the MD5 class
6916       klass_name = "sun/security/provider/MD5";
6917     }
6918     break;
6919   case 1:
6920     if (UseSHA1Intrinsics) {
6921       // we want to do an instanceof comparison against the SHA class
6922       klass_name = "sun/security/provider/SHA";
6923     }
6924     break;
6925   case 2:
6926     if (UseSHA256Intrinsics) {
6927       // we want to do an instanceof comparison against the SHA2 class
6928       klass_name = "sun/security/provider/SHA2";
6929     }
6930     break;
6931   case 3:
6932     if (UseSHA512Intrinsics) {
6933       // we want to do an instanceof comparison against the SHA5 class
6934       klass_name = "sun/security/provider/SHA5";
6935     }
6936     break;
6937   case 4:
6938     if (UseSHA3Intrinsics) {
6939       // we want to do an instanceof comparison against the SHA3 class
6940       klass_name = "sun/security/provider/SHA3";
6941     }
6942     break;
6943   default:
6944     fatal("unknown SHA intrinsic predicate: %d", predicate);
6945   }
6946 
6947   ciKlass* klass = NULL;
6948   if (klass_name != NULL) {
6949     klass = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make(klass_name));
6950   }
6951   if ((klass == NULL) || !klass->is_loaded()) {
6952     // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
6953     Node* ctrl = control();
6954     set_control(top()); // no intrinsic path
6955     return ctrl;
6956   }
6957   ciInstanceKlass* instklass = klass->as_instance_klass();
6958 
6959   Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
6960   Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
6961   Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
6962   Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);
6963 
6964   return instof_false;  // even if it is NULL
6965 }
6966 
6967 //-------------inline_fma-----------------------------------
6968 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
6969   Node *a = NULL;
6970   Node *b = NULL;
6971   Node *c = NULL;
6972   Node* result = NULL;
6973   switch (id) {
6974   case vmIntrinsics::_fmaD:
6975     assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
6976     // no receiver since it is static method
6977     a = round_double_node(argument(0));
6978     b = round_double_node(argument(2));
6979     c = round_double_node(argument(4));
6980     result = _gvn.transform(new FmaDNode(control(), a, b, c));
6981     break;
6982   case vmIntrinsics::_fmaF:
6983     assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
6984     a = argument(0);
6985     b = argument(1);
6986     c = argument(2);
6987     result = _gvn.transform(new FmaFNode(control(), a, b, c));
6988     break;
6989   default:
6990     fatal_unexpected_iid(id);  break;
6991   }
6992   set_result(result);
6993   return true;
6994 }
6995 
6996 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
6997   // argument(0) is receiver
6998   Node* codePoint = argument(1);
6999   Node* n = NULL;
7000 
7001   switch (id) {
7002     case vmIntrinsics::_isDigit :
7003       n = new DigitNode(control(), codePoint);
7004       break;
7005     case vmIntrinsics::_isLowerCase :
7006       n = new LowerCaseNode(control(), codePoint);
7007       break;
7008     case vmIntrinsics::_isUpperCase :
7009       n = new UpperCaseNode(control(), codePoint);
7010       break;
7011     case vmIntrinsics::_isWhitespace :
7012       n = new WhitespaceNode(control(), codePoint);
7013       break;
7014     default:
7015       fatal_unexpected_iid(id);
7016   }
7017 
7018   set_result(_gvn.transform(n));
7019   return true;
7020 }
7021 
7022 //------------------------------inline_fp_min_max------------------------------
7023 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
7024 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
7025 
7026   // The intrinsic should be used only when the API branches aren't predictable,
7027   // the last one performing the most important comparison. The following heuristic
7028   // uses the branch statistics to eventually bail out if necessary.
7029 
7030   ciMethodData *md = callee()->method_data();
7031 
7032   if ( md != NULL && md->is_mature() && md->invocation_count() > 0 ) {
7033     ciCallProfile cp = caller()->call_profile_at_bci(bci());
7034 
7035     if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
7036       // Bail out if the call-site didn't contribute enough to the statistics.
7037       return false;
7038     }
7039 
7040     uint taken = 0, not_taken = 0;
7041 
7042     for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
7043       if (p->is_BranchData()) {
7044         taken = ((ciBranchData*)p)->taken();
7045         not_taken = ((ciBranchData*)p)->not_taken();
7046       }
7047     }
7048 
7049     double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
7050     balance = balance < 0 ? -balance : balance;
7051     if ( balance > 0.2 ) {
7052       // Bail out if the most important branch is predictable enough.
7053       return false;
7054     }
7055   }
7056 */
7057 
7058   Node *a = NULL;
7059   Node *b = NULL;
7060   Node *n = NULL;
7061   switch (id) {
7062   case vmIntrinsics::_maxF:
7063   case vmIntrinsics::_minF:
7064   case vmIntrinsics::_maxF_strict:
7065   case vmIntrinsics::_minF_strict:
7066     assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
7067     a = argument(0);
7068     b = argument(1);
7069     break;
7070   case vmIntrinsics::_maxD:
7071   case vmIntrinsics::_minD:
7072   case vmIntrinsics::_maxD_strict:
7073   case vmIntrinsics::_minD_strict:
7074     assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
7075     a = round_double_node(argument(0));
7076     b = round_double_node(argument(2));
7077     break;
7078   default:
7079     fatal_unexpected_iid(id);
7080     break;
7081   }
7082   switch (id) {
7083   case vmIntrinsics::_maxF:
7084   case vmIntrinsics::_maxF_strict:
7085     n = new MaxFNode(a, b);
7086     break;
7087   case vmIntrinsics::_minF:
7088   case vmIntrinsics::_minF_strict:
7089     n = new MinFNode(a, b);
7090     break;
7091   case vmIntrinsics::_maxD:
7092   case vmIntrinsics::_maxD_strict:
7093     n = new MaxDNode(a, b);
7094     break;
7095   case vmIntrinsics::_minD:
7096   case vmIntrinsics::_minD_strict:
7097     n = new MinDNode(a, b);
7098     break;
7099   default:
7100     fatal_unexpected_iid(id);
7101     break;
7102   }
7103   set_result(_gvn.transform(n));
7104   return true;
7105 }
7106 
7107 bool LibraryCallKit::inline_profileBoolean() {
7108   Node* counts = argument(1);
7109   const TypeAryPtr* ary = NULL;
7110   ciArray* aobj = NULL;
7111   if (counts->is_Con()
7112       && (ary = counts->bottom_type()->isa_aryptr()) != NULL
7113       && (aobj = ary->const_oop()->as_array()) != NULL
7114       && (aobj->length() == 2)) {
7115     // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
7116     jint false_cnt = aobj->element_value(0).as_int();
7117     jint  true_cnt = aobj->element_value(1).as_int();
7118 
7119     if (C->log() != NULL) {
7120       C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
7121                      false_cnt, true_cnt);
7122     }
7123 
7124     if (false_cnt + true_cnt == 0) {
7125       // According to profile, never executed.
7126       uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7127                           Deoptimization::Action_reinterpret);
7128       return true;
7129     }
7130 
7131     // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
7132     // is a number of each value occurrences.
7133     Node* result = argument(0);
7134     if (false_cnt == 0 || true_cnt == 0) {
7135       // According to profile, one value has been never seen.
7136       int expected_val = (false_cnt == 0) ? 1 : 0;
7137 
7138       Node* cmp  = _gvn.transform(new CmpINode(result, intcon(expected_val)));
7139       Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7140 
7141       IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
7142       Node* fast_path = _gvn.transform(new IfTrueNode(check));
7143       Node* slow_path = _gvn.transform(new IfFalseNode(check));
7144 
7145       { // Slow path: uncommon trap for never seen value and then reexecute
7146         // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
7147         // the value has been seen at least once.
7148         PreserveJVMState pjvms(this);
7149         PreserveReexecuteState preexecs(this);
7150         jvms()->set_should_reexecute(true);
7151 
7152         set_control(slow_path);
7153         set_i_o(i_o());
7154 
7155         uncommon_trap_exact(Deoptimization::Reason_intrinsic,
7156                             Deoptimization::Action_reinterpret);
7157       }
7158       // The guard for never seen value enables sharpening of the result and
7159       // returning a constant. It allows to eliminate branches on the same value
7160       // later on.
7161       set_control(fast_path);
7162       result = intcon(expected_val);
7163     }
7164     // Stop profiling.
7165     // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
7166     // By replacing method body with profile data (represented as ProfileBooleanNode
7167     // on IR level) we effectively disable profiling.
7168     // It enables full speed execution once optimized code is generated.
7169     Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
7170     C->record_for_igvn(profile);
7171     set_result(profile);
7172     return true;
7173   } else {
7174     // Continue profiling.
7175     // Profile data isn't available at the moment. So, execute method's bytecode version.
7176     // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
7177     // is compiled and counters aren't available since corresponding MethodHandle
7178     // isn't a compile-time constant.
7179     return false;
7180   }
7181 }
7182 
7183 bool LibraryCallKit::inline_isCompileConstant() {
7184   Node* n = argument(0);
7185   set_result(n->is_Con() ? intcon(1) : intcon(0));
7186   return true;
7187 }
7188 
7189 //------------------------------- inline_getObjectSize --------------------------------------
7190 //
7191 // Calculate the runtime size of the object/array.
7192 //   native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
7193 //
7194 bool LibraryCallKit::inline_getObjectSize() {
7195   Node* obj = argument(3);
7196   Node* klass_node = load_object_klass(obj);
7197 
7198   jint  layout_con = Klass::_lh_neutral_value;
7199   Node* layout_val = get_layout_helper(klass_node, layout_con);
7200   int   layout_is_con = (layout_val == NULL);
7201 
7202   if (layout_is_con) {
7203     // Layout helper is constant, can figure out things at compile time.
7204 
7205     if (Klass::layout_helper_is_instance(layout_con)) {
7206       // Instance case:  layout_con contains the size itself.
7207       Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
7208       set_result(size);
7209     } else {
7210       // Array case: size is round(header + element_size*arraylength).
7211       // Since arraylength is different for every array instance, we have to
7212       // compute the whole thing at runtime.
7213 
7214       Node* arr_length = load_array_length(obj);
7215 
7216       int round_mask = MinObjAlignmentInBytes - 1;
7217       int hsize  = Klass::layout_helper_header_size(layout_con);
7218       int eshift = Klass::layout_helper_log2_element_size(layout_con);
7219 
7220       if ((round_mask & ~right_n_bits(eshift)) == 0) {
7221         round_mask = 0;  // strength-reduce it if it goes away completely
7222       }
7223       assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
7224       Node* header_size = intcon(hsize + round_mask);
7225 
7226       Node* lengthx = ConvI2X(arr_length);
7227       Node* headerx = ConvI2X(header_size);
7228 
7229       Node* abody = lengthx;
7230       if (eshift != 0) {
7231         abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
7232       }
7233       Node* size = _gvn.transform( new AddXNode(headerx, abody) );
7234       if (round_mask != 0) {
7235         size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
7236       }
7237       size = ConvX2L(size);
7238       set_result(size);
7239     }
7240   } else {
7241     // Layout helper is not constant, need to test for array-ness at runtime.
7242 
7243     enum { _instance_path = 1, _array_path, PATH_LIMIT };
7244     RegionNode* result_reg = new RegionNode(PATH_LIMIT);
7245     PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
7246     record_for_igvn(result_reg);
7247 
7248     Node* array_ctl = generate_array_guard(klass_node, NULL);
7249     if (array_ctl != NULL) {
7250       // Array case: size is round(header + element_size*arraylength).
7251       // Since arraylength is different for every array instance, we have to
7252       // compute the whole thing at runtime.
7253 
7254       PreserveJVMState pjvms(this);
7255       set_control(array_ctl);
7256       Node* arr_length = load_array_length(obj);
7257 
7258       int round_mask = MinObjAlignmentInBytes - 1;
7259       Node* mask = intcon(round_mask);
7260 
7261       Node* hss = intcon(Klass::_lh_header_size_shift);
7262       Node* hsm = intcon(Klass::_lh_header_size_mask);
7263       Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
7264       header_size = _gvn.transform(new AndINode(header_size, hsm));
7265       header_size = _gvn.transform(new AddINode(header_size, mask));
7266 
7267       // There is no need to mask or shift this value.
7268       // The semantics of LShiftINode include an implicit mask to 0x1F.
7269       assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
7270       Node* elem_shift = layout_val;
7271 
7272       Node* lengthx = ConvI2X(arr_length);
7273       Node* headerx = ConvI2X(header_size);
7274 
7275       Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
7276       Node* size = _gvn.transform(new AddXNode(headerx, abody));
7277       if (round_mask != 0) {
7278         size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
7279       }
7280       size = ConvX2L(size);
7281 
7282       result_reg->init_req(_array_path, control());
7283       result_val->init_req(_array_path, size);
7284     }
7285 
7286     if (!stopped()) {
7287       // Instance case: the layout helper gives us instance size almost directly,
7288       // but we need to mask out the _lh_instance_slow_path_bit.
7289       Node* size = ConvI2X(layout_val);
7290       assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
7291       Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
7292       size = _gvn.transform(new AndXNode(size, mask));
7293       size = ConvX2L(size);
7294 
7295       result_reg->init_req(_instance_path, control());
7296       result_val->init_req(_instance_path, size);
7297     }
7298 
7299     set_result(result_reg, result_val);
7300   }
7301 
7302   return true;
7303 }
7304 
7305 //------------------------------- inline_blackhole --------------------------------------
7306 //
7307 // Make sure all arguments to this node are alive.
7308 // This matches methods that were requested to be blackholed through compile commands.
7309 //
7310 bool LibraryCallKit::inline_blackhole() {
7311   assert(callee()->is_static(), "Should have been checked before: only static methods here");
7312   assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
7313   assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
7314 
7315   // Bind call arguments as blackhole arguments to keep them alive
7316   Node* bh = insert_mem_bar(Op_Blackhole);
7317   uint nargs = callee()->arg_size();
7318   for (uint i = 0; i < nargs; i++) {
7319     bh->add_req(argument(i));
7320   }
7321 
7322   return true;
7323 }