library_call.cpp 248.4 KB
Newer Older
D
duke 已提交
1
/*
S
sla 已提交
2
 * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
D
duke 已提交
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
19 20 21
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
D
duke 已提交
22 23 24
 *
 */

25 26 27
#include "precompiled.hpp"
#include "classfile/systemDictionary.hpp"
#include "classfile/vmSymbols.hpp"
28
#include "compiler/compileBroker.hpp"
29 30 31 32 33 34
#include "compiler/compileLog.hpp"
#include "oops/objArrayKlass.hpp"
#include "opto/addnode.hpp"
#include "opto/callGenerator.hpp"
#include "opto/cfgnode.hpp"
#include "opto/idealKit.hpp"
35
#include "opto/mathexactnode.hpp"
36 37 38 39 40 41
#include "opto/mulnode.hpp"
#include "opto/parse.hpp"
#include "opto/runtime.hpp"
#include "opto/subnode.hpp"
#include "prims/nativeLookup.hpp"
#include "runtime/sharedRuntime.hpp"
S
sla 已提交
42
#include "trace/traceMacros.hpp"
D
duke 已提交
43 44 45 46 47 48

class LibraryIntrinsic : public InlineCallGenerator {
  // Extend the set of intrinsics known to the runtime:
 public:
 private:
  bool             _is_virtual;
49
  bool             _is_predicted;
50
  bool             _does_virtual_dispatch;
D
duke 已提交
51 52 53
  vmIntrinsics::ID _intrinsic_id;

 public:
54
  LibraryIntrinsic(ciMethod* m, bool is_virtual, bool is_predicted, bool does_virtual_dispatch, vmIntrinsics::ID id)
D
duke 已提交
55 56
    : InlineCallGenerator(m),
      _is_virtual(is_virtual),
57
      _is_predicted(is_predicted),
58
      _does_virtual_dispatch(does_virtual_dispatch),
D
duke 已提交
59 60 61 62 63
      _intrinsic_id(id)
  {
  }
  virtual bool is_intrinsic() const { return true; }
  virtual bool is_virtual()   const { return _is_virtual; }
64
  virtual bool is_predicted()   const { return _is_predicted; }
65
  virtual bool does_virtual_dispatch()   const { return _does_virtual_dispatch; }
66
  virtual JVMState* generate(JVMState* jvms, Parse* parent_parser);
67
  virtual Node* generate_predicate(JVMState* jvms);
D
duke 已提交
68 69 70 71 72 73 74
  vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
};


// Local helper class for LibraryIntrinsic:
class LibraryCallKit : public GraphKit {
 private:
75 76 77
  LibraryIntrinsic* _intrinsic;     // the library intrinsic being called
  Node*             _result;        // the result node, if any
  int               _reexecute_sp;  // the stack pointer when bytecode needs to be reexecuted
D
duke 已提交
78

79 80
  const TypeOopPtr* sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr = false);

D
duke 已提交
81
 public:
82 83 84 85
  LibraryCallKit(JVMState* jvms, LibraryIntrinsic* intrinsic)
    : GraphKit(jvms),
      _intrinsic(intrinsic),
      _result(NULL)
D
duke 已提交
86
  {
87 88 89 90 91 92 93 94 95 96 97 98 99
    // Check if this is a root compile.  In that case we don't have a caller.
    if (!jvms->has_method()) {
      _reexecute_sp = sp();
    } else {
      // Find out how many arguments the interpreter needs when deoptimizing
      // and save the stack pointer value so it can used by uncommon_trap.
      // We find the argument count by looking at the declared signature.
      bool ignored_will_link;
      ciSignature* declared_signature = NULL;
      ciMethod* ignored_callee = caller()->get_method_at_bci(bci(), ignored_will_link, &declared_signature);
      const int nargs = declared_signature->arg_size_for_bc(caller()->java_code_at_bci(bci()));
      _reexecute_sp = sp() + nargs;  // "push" arguments back on stack
    }
D
duke 已提交
100 101
  }

102 103
  virtual LibraryCallKit* is_LibraryCallKit() const { return (LibraryCallKit*)this; }

D
duke 已提交
104 105 106 107 108 109 110
  ciMethod*         caller()    const    { return jvms()->method(); }
  int               bci()       const    { return jvms()->bci(); }
  LibraryIntrinsic* intrinsic() const    { return _intrinsic; }
  vmIntrinsics::ID  intrinsic_id() const { return _intrinsic->intrinsic_id(); }
  ciMethod*         callee()    const    { return _intrinsic->method(); }

  bool try_to_inline();
111
  Node* try_to_predicate();
D
duke 已提交
112

113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131
  void push_result() {
    // Push the result onto the stack.
    if (!stopped() && result() != NULL) {
      BasicType bt = result()->bottom_type()->basic_type();
      push_node(bt, result());
    }
  }

 private:
  void fatal_unexpected_iid(vmIntrinsics::ID iid) {
    fatal(err_msg_res("unexpected intrinsic %d: %s", iid, vmIntrinsics::name_at(iid)));
  }

  void  set_result(Node* n) { assert(_result == NULL, "only set once"); _result = n; }
  void  set_result(RegionNode* region, PhiNode* value);
  Node*     result() { return _result; }

  virtual int reexecute_sp() { return _reexecute_sp; }

D
duke 已提交
132 133 134 135 136 137 138 139 140 141 142 143 144 145 146
  // Helper functions to inline natives
  Node* generate_guard(Node* test, RegionNode* region, float true_prob);
  Node* generate_slow_guard(Node* test, RegionNode* region);
  Node* generate_fair_guard(Node* test, RegionNode* region);
  Node* generate_negative_guard(Node* index, RegionNode* region,
                                // resulting CastII of index:
                                Node* *pos_index = NULL);
  Node* generate_nonpositive_guard(Node* index, bool never_negative,
                                   // resulting CastII of index:
                                   Node* *pos_index = NULL);
  Node* generate_limit_guard(Node* offset, Node* subseq_length,
                             Node* array_length,
                             RegionNode* region);
  Node* generate_current_thread(Node* &tls_output);
  address basictype2arraycopy(BasicType t, Node *src_offset, Node *dest_offset,
147
                              bool disjoint_bases, const char* &name, bool dest_uninitialized);
D
duke 已提交
148 149 150 151
  Node* load_mirror_from_klass(Node* klass);
  Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
                                      RegionNode* region, int null_path,
                                      int offset);
152
  Node* load_klass_from_mirror(Node* mirror, bool never_see_null,
D
duke 已提交
153 154
                               RegionNode* region, int null_path) {
    int offset = java_lang_Class::klass_offset_in_bytes();
155
    return load_klass_from_mirror_common(mirror, never_see_null,
D
duke 已提交
156 157 158 159 160 161
                                         region, null_path,
                                         offset);
  }
  Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
                                     RegionNode* region, int null_path) {
    int offset = java_lang_Class::array_klass_offset_in_bytes();
162
    return load_klass_from_mirror_common(mirror, never_see_null,
D
duke 已提交
163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
                                         region, null_path,
                                         offset);
  }
  Node* generate_access_flags_guard(Node* kls,
                                    int modifier_mask, int modifier_bits,
                                    RegionNode* region);
  Node* generate_interface_guard(Node* kls, RegionNode* region);
  Node* generate_array_guard(Node* kls, RegionNode* region) {
    return generate_array_guard_common(kls, region, false, false);
  }
  Node* generate_non_array_guard(Node* kls, RegionNode* region) {
    return generate_array_guard_common(kls, region, false, true);
  }
  Node* generate_objArray_guard(Node* kls, RegionNode* region) {
    return generate_array_guard_common(kls, region, true, false);
  }
  Node* generate_non_objArray_guard(Node* kls, RegionNode* region) {
    return generate_array_guard_common(kls, region, true, true);
  }
  Node* generate_array_guard_common(Node* kls, RegionNode* region,
                                    bool obj_array, bool not_array);
  Node* generate_virtual_guard(Node* obj_klass, RegionNode* slow_region);
  CallJavaNode* generate_method_call(vmIntrinsics::ID method_id,
                                     bool is_virtual = false, bool is_static = false);
  CallJavaNode* generate_method_call_static(vmIntrinsics::ID method_id) {
    return generate_method_call(method_id, false, true);
  }
  CallJavaNode* generate_method_call_virtual(vmIntrinsics::ID method_id) {
    return generate_method_call(method_id, true, false);
  }
193
  Node * load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString, bool is_exact, bool is_static);
D
duke 已提交
194

K
kvn 已提交
195 196
  Node* make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2);
  Node* make_string_method_node(int opcode, Node* str1, Node* str2);
D
duke 已提交
197 198 199
  bool inline_string_compareTo();
  bool inline_string_indexOf();
  Node* string_indexOf(Node* string_object, ciTypeArray* target_array, jint offset, jint cache_i, jint md2_i);
C
cfang 已提交
200
  bool inline_string_equals();
201
  Node* round_double_node(Node* n);
D
duke 已提交
202 203 204
  bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
  bool inline_math_native(vmIntrinsics::ID id);
  bool inline_trig(vmIntrinsics::ID id);
205
  bool inline_math(vmIntrinsics::ID id);
206 207 208 209 210 211 212 213 214
  void inline_math_mathExact(Node* math);
  bool inline_math_addExactI(bool is_increment);
  bool inline_math_addExactL(bool is_increment);
  bool inline_math_multiplyExactI();
  bool inline_math_multiplyExactL();
  bool inline_math_negateExactI();
  bool inline_math_negateExactL();
  bool inline_math_subtractExactI(bool is_decrement);
  bool inline_math_subtractExactL(bool is_decrement);
215 216
  bool inline_exp();
  bool inline_pow();
217
  void finish_pow_exp(Node* result, Node* x, Node* y, const TypeFunc* call_type, address funcAddr, const char* funcName);
D
duke 已提交
218 219 220 221 222
  bool inline_min_max(vmIntrinsics::ID id);
  Node* generate_min_max(vmIntrinsics::ID id, Node* x, Node* y);
  // This returns Type::AnyPtr, RawPtr, or OopPtr.
  int classify_unsafe_addr(Node* &base, Node* &offset);
  Node* make_unsafe_address(Node* base, Node* offset);
223 224 225
  // Helper for inline_unsafe_access.
  // Generates the guards that check whether the result of
  // Unsafe.getObject should be recorded in an SATB log buffer.
226
  void insert_pre_barrier(Node* base_oop, Node* offset, Node* pre_val, bool need_mem_bar);
D
duke 已提交
227 228
  bool inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, bool is_volatile);
  bool inline_unsafe_prefetch(bool is_native_ptr, bool is_store, bool is_static);
R
rbackman 已提交
229
  static bool klass_needs_init_guard(Node* kls);
D
duke 已提交
230 231 232
  bool inline_unsafe_allocate();
  bool inline_unsafe_copyMemory();
  bool inline_native_currentThread();
233 234 235 236 237
#ifdef TRACE_HAVE_INTRINSICS
  bool inline_native_classID();
  bool inline_native_threadID();
#endif
  bool inline_native_time_funcs(address method, const char* funcName);
D
duke 已提交
238 239 240 241 242 243 244
  bool inline_native_isInterrupted();
  bool inline_native_Class_query(vmIntrinsics::ID id);
  bool inline_native_subtype_check();

  bool inline_native_newArray();
  bool inline_native_getLength();
  bool inline_array_copyOf(bool is_copyOfRange);
245
  bool inline_array_equals();
246
  void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
D
duke 已提交
247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275
  bool inline_native_clone(bool is_virtual);
  bool inline_native_Reflection_getCallerClass();
  // Helper function for inlining native object hash method
  bool inline_native_hashcode(bool is_virtual, bool is_static);
  bool inline_native_getClass();

  // Helper functions for inlining arraycopy
  bool inline_arraycopy();
  void generate_arraycopy(const TypePtr* adr_type,
                          BasicType basic_elem_type,
                          Node* src,  Node* src_offset,
                          Node* dest, Node* dest_offset,
                          Node* copy_length,
                          bool disjoint_bases = false,
                          bool length_never_negative = false,
                          RegionNode* slow_region = NULL);
  AllocateArrayNode* tightly_coupled_allocation(Node* ptr,
                                                RegionNode* slow_region);
  void generate_clear_array(const TypePtr* adr_type,
                            Node* dest,
                            BasicType basic_elem_type,
                            Node* slice_off,
                            Node* slice_len,
                            Node* slice_end);
  bool generate_block_arraycopy(const TypePtr* adr_type,
                                BasicType basic_elem_type,
                                AllocateNode* alloc,
                                Node* src,  Node* src_offset,
                                Node* dest, Node* dest_offset,
276
                                Node* dest_size, bool dest_uninitialized);
D
duke 已提交
277 278 279
  void generate_slow_arraycopy(const TypePtr* adr_type,
                               Node* src,  Node* src_offset,
                               Node* dest, Node* dest_offset,
280
                               Node* copy_length, bool dest_uninitialized);
D
duke 已提交
281 282 283 284
  Node* generate_checkcast_arraycopy(const TypePtr* adr_type,
                                     Node* dest_elem_klass,
                                     Node* src,  Node* src_offset,
                                     Node* dest, Node* dest_offset,
285
                                     Node* copy_length, bool dest_uninitialized);
D
duke 已提交
286 287 288
  Node* generate_generic_arraycopy(const TypePtr* adr_type,
                                   Node* src,  Node* src_offset,
                                   Node* dest, Node* dest_offset,
289
                                   Node* copy_length, bool dest_uninitialized);
D
duke 已提交
290 291 292 293 294
  void generate_unchecked_arraycopy(const TypePtr* adr_type,
                                    BasicType basic_elem_type,
                                    bool disjoint_bases,
                                    Node* src,  Node* src_offset,
                                    Node* dest, Node* dest_offset,
295
                                    Node* copy_length, bool dest_uninitialized);
296 297
  typedef enum { LS_xadd, LS_xchg, LS_cmpxchg } LoadStoreKind;
  bool inline_unsafe_load_store(BasicType type,  LoadStoreKind kind);
D
duke 已提交
298
  bool inline_unsafe_ordered_store(BasicType type);
299
  bool inline_unsafe_fence(vmIntrinsics::ID id);
D
duke 已提交
300
  bool inline_fp_conversions(vmIntrinsics::ID id);
301
  bool inline_number_methods(vmIntrinsics::ID id);
302
  bool inline_reference_get();
303 304 305 306
  bool inline_aescrypt_Block(vmIntrinsics::ID id);
  bool inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id);
  Node* inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting);
  Node* get_key_start_from_aescrypt_object(Node* aescrypt_object);
307
  bool inline_encodeISOArray();
308 309 310
  bool inline_updateCRC32();
  bool inline_updateBytesCRC32();
  bool inline_updateByteBufferCRC32();
D
duke 已提交
311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
};


//---------------------------make_vm_intrinsic----------------------------
CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
  vmIntrinsics::ID id = m->intrinsic_id();
  assert(id != vmIntrinsics::_none, "must be a VM intrinsic");

  if (DisableIntrinsic[0] != '\0'
      && strstr(DisableIntrinsic, vmIntrinsics::name_at(id)) != NULL) {
    // disabled by a user request on the command line:
    // example: -XX:DisableIntrinsic=_hashCode,_getClass
    return NULL;
  }

  if (!m->is_loaded()) {
    // do not attempt to inline unloaded methods
    return NULL;
  }

  // Only a few intrinsics implement a virtual dispatch.
  // They are expensive calls which are also frequently overridden.
  if (is_virtual) {
    switch (id) {
    case vmIntrinsics::_hashCode:
    case vmIntrinsics::_clone:
      // OK, Object.hashCode and Object.clone intrinsics come in both flavors
      break;
    default:
      return NULL;
    }
  }

  // -XX:-InlineNatives disables nearly all intrinsics:
  if (!InlineNatives) {
    switch (id) {
    case vmIntrinsics::_indexOf:
    case vmIntrinsics::_compareTo:
C
cfang 已提交
349
    case vmIntrinsics::_equals:
350
    case vmIntrinsics::_equalsC:
351 352 353 354 355
    case vmIntrinsics::_getAndAddInt:
    case vmIntrinsics::_getAndAddLong:
    case vmIntrinsics::_getAndSetInt:
    case vmIntrinsics::_getAndSetLong:
    case vmIntrinsics::_getAndSetObject:
356 357 358
    case vmIntrinsics::_loadFence:
    case vmIntrinsics::_storeFence:
    case vmIntrinsics::_fullFence:
D
duke 已提交
359
      break;  // InlineNatives does not control String.compareTo
360 361
    case vmIntrinsics::_Reference_get:
      break;  // InlineNatives does not control Reference.get
D
duke 已提交
362 363 364 365 366
    default:
      return NULL;
    }
  }

367
  bool is_predicted = false;
368
  bool does_virtual_dispatch = false;
369

D
duke 已提交
370 371 372
  switch (id) {
  case vmIntrinsics::_compareTo:
    if (!SpecialStringCompareTo)  return NULL;
373
    if (!Matcher::match_rule_supported(Op_StrComp))  return NULL;
D
duke 已提交
374 375 376 377
    break;
  case vmIntrinsics::_indexOf:
    if (!SpecialStringIndexOf)  return NULL;
    break;
C
cfang 已提交
378 379
  case vmIntrinsics::_equals:
    if (!SpecialStringEquals)  return NULL;
380
    if (!Matcher::match_rule_supported(Op_StrEquals))  return NULL;
C
cfang 已提交
381
    break;
382 383
  case vmIntrinsics::_equalsC:
    if (!SpecialArraysEquals)  return NULL;
384
    if (!Matcher::match_rule_supported(Op_AryEq))  return NULL;
385
    break;
D
duke 已提交
386 387 388 389 390 391 392 393 394
  case vmIntrinsics::_arraycopy:
    if (!InlineArrayCopy)  return NULL;
    break;
  case vmIntrinsics::_copyMemory:
    if (StubRoutines::unsafe_arraycopy() == NULL)  return NULL;
    if (!InlineArrayCopy)  return NULL;
    break;
  case vmIntrinsics::_hashCode:
    if (!InlineObjectHash)  return NULL;
395
    does_virtual_dispatch = true;
D
duke 已提交
396 397
    break;
  case vmIntrinsics::_clone:
398
    does_virtual_dispatch = true;
D
duke 已提交
399 400 401 402 403 404
  case vmIntrinsics::_copyOf:
  case vmIntrinsics::_copyOfRange:
    if (!InlineObjectCopy)  return NULL;
    // These also use the arraycopy intrinsic mechanism:
    if (!InlineArrayCopy)  return NULL;
    break;
405 406 407 408
  case vmIntrinsics::_encodeISOArray:
    if (!SpecialEncodeISOArray)  return NULL;
    if (!Matcher::match_rule_supported(Op_EncodeISOArray))  return NULL;
    break;
D
duke 已提交
409 410 411 412 413 414 415
  case vmIntrinsics::_checkIndex:
    // We do not intrinsify this.  The optimizer does fine with it.
    return NULL;

  case vmIntrinsics::_getCallerClass:
    if (!UseNewReflection)  return NULL;
    if (!InlineReflectionGetCallerClass)  return NULL;
416
    if (SystemDictionary::reflect_CallerSensitive_klass() == NULL)  return NULL;
D
duke 已提交
417 418
    break;

419
  case vmIntrinsics::_bitCount_i:
420
    if (!Matcher::match_rule_supported(Op_PopCountI)) return NULL;
421 422
    break;

423
  case vmIntrinsics::_bitCount_l:
424
    if (!Matcher::match_rule_supported(Op_PopCountL)) return NULL;
425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440
    break;

  case vmIntrinsics::_numberOfLeadingZeros_i:
    if (!Matcher::match_rule_supported(Op_CountLeadingZerosI)) return NULL;
    break;

  case vmIntrinsics::_numberOfLeadingZeros_l:
    if (!Matcher::match_rule_supported(Op_CountLeadingZerosL)) return NULL;
    break;

  case vmIntrinsics::_numberOfTrailingZeros_i:
    if (!Matcher::match_rule_supported(Op_CountTrailingZerosI)) return NULL;
    break;

  case vmIntrinsics::_numberOfTrailingZeros_l:
    if (!Matcher::match_rule_supported(Op_CountTrailingZerosL)) return NULL;
441 442
    break;

443
  case vmIntrinsics::_reverseBytes_c:
444
    if (!Matcher::match_rule_supported(Op_ReverseBytesUS)) return NULL;
445 446
    break;
  case vmIntrinsics::_reverseBytes_s:
447
    if (!Matcher::match_rule_supported(Op_ReverseBytesS))  return NULL;
448 449
    break;
  case vmIntrinsics::_reverseBytes_i:
450
    if (!Matcher::match_rule_supported(Op_ReverseBytesI))  return NULL;
451 452
    break;
  case vmIntrinsics::_reverseBytes_l:
453
    if (!Matcher::match_rule_supported(Op_ReverseBytesL))  return NULL;
454 455
    break;

456
  case vmIntrinsics::_Reference_get:
457 458 459 460
    // Use the intrinsic version of Reference.get() so that the value in
    // the referent field can be registered by the G1 pre-barrier code.
    // Also add memory barrier to prevent commoning reads from this field
    // across safepoint since GC can change it value.
461 462
    break;

463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
  case vmIntrinsics::_compareAndSwapObject:
#ifdef _LP64
    if (!UseCompressedOops && !Matcher::match_rule_supported(Op_CompareAndSwapP)) return NULL;
#endif
    break;

  case vmIntrinsics::_compareAndSwapLong:
    if (!Matcher::match_rule_supported(Op_CompareAndSwapL)) return NULL;
    break;

  case vmIntrinsics::_getAndAddInt:
    if (!Matcher::match_rule_supported(Op_GetAndAddI)) return NULL;
    break;

  case vmIntrinsics::_getAndAddLong:
    if (!Matcher::match_rule_supported(Op_GetAndAddL)) return NULL;
    break;

  case vmIntrinsics::_getAndSetInt:
    if (!Matcher::match_rule_supported(Op_GetAndSetI)) return NULL;
    break;

  case vmIntrinsics::_getAndSetLong:
    if (!Matcher::match_rule_supported(Op_GetAndSetL)) return NULL;
    break;

  case vmIntrinsics::_getAndSetObject:
#ifdef _LP64
    if (!UseCompressedOops && !Matcher::match_rule_supported(Op_GetAndSetP)) return NULL;
    if (UseCompressedOops && !Matcher::match_rule_supported(Op_GetAndSetN)) return NULL;
    break;
#else
    if (!Matcher::match_rule_supported(Op_GetAndSetP)) return NULL;
    break;
#endif

499 500 501 502 503 504 505 506 507 508 509 510
  case vmIntrinsics::_aescrypt_encryptBlock:
  case vmIntrinsics::_aescrypt_decryptBlock:
    if (!UseAESIntrinsics) return NULL;
    break;

  case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
  case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
    if (!UseAESIntrinsics) return NULL;
    // these two require the predicated logic
    is_predicted = true;
    break;

511 512 513 514 515 516
  case vmIntrinsics::_updateCRC32:
  case vmIntrinsics::_updateBytesCRC32:
  case vmIntrinsics::_updateByteBufferCRC32:
    if (!UseCRC32Intrinsics) return NULL;
    break;

517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543
  case vmIntrinsics::_incrementExactI:
  case vmIntrinsics::_addExactI:
    if (!Matcher::match_rule_supported(Op_AddExactI) || !UseMathExactIntrinsics) return NULL;
    break;
  case vmIntrinsics::_incrementExactL:
  case vmIntrinsics::_addExactL:
    if (!Matcher::match_rule_supported(Op_AddExactL) || !UseMathExactIntrinsics) return NULL;
    break;
  case vmIntrinsics::_decrementExactI:
  case vmIntrinsics::_subtractExactI:
    if (!Matcher::match_rule_supported(Op_SubExactI) || !UseMathExactIntrinsics) return NULL;
    break;
  case vmIntrinsics::_decrementExactL:
  case vmIntrinsics::_subtractExactL:
    if (!Matcher::match_rule_supported(Op_SubExactL) || !UseMathExactIntrinsics) return NULL;
    break;
  case vmIntrinsics::_negateExactI:
    if (!Matcher::match_rule_supported(Op_NegExactI) || !UseMathExactIntrinsics) return NULL;
    break;
  case vmIntrinsics::_negateExactL:
    if (!Matcher::match_rule_supported(Op_NegExactL) || !UseMathExactIntrinsics) return NULL;
    break;
  case vmIntrinsics::_multiplyExactI:
    if (!Matcher::match_rule_supported(Op_MulExactI) || !UseMathExactIntrinsics) return NULL;
    break;
  case vmIntrinsics::_multiplyExactL:
    if (!Matcher::match_rule_supported(Op_MulExactL) || !UseMathExactIntrinsics) return NULL;
544 545
    break;

D
duke 已提交
546
 default:
547 548
    assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
    assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
D
duke 已提交
549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576
    break;
  }

  // -XX:-InlineClassNatives disables natives from the Class class.
  // The flag applies to all reflective calls, notably Array.newArray
  // (visible to Java programmers as Array.newInstance).
  if (m->holder()->name() == ciSymbol::java_lang_Class() ||
      m->holder()->name() == ciSymbol::java_lang_reflect_Array()) {
    if (!InlineClassNatives)  return NULL;
  }

  // -XX:-InlineThreadNatives disables natives from the Thread class.
  if (m->holder()->name() == ciSymbol::java_lang_Thread()) {
    if (!InlineThreadNatives)  return NULL;
  }

  // -XX:-InlineMathNatives disables natives from the Math,Float and Double classes.
  if (m->holder()->name() == ciSymbol::java_lang_Math() ||
      m->holder()->name() == ciSymbol::java_lang_Float() ||
      m->holder()->name() == ciSymbol::java_lang_Double()) {
    if (!InlineMathNatives)  return NULL;
  }

  // -XX:-InlineUnsafeOps disables natives from the Unsafe class.
  if (m->holder()->name() == ciSymbol::sun_misc_Unsafe()) {
    if (!InlineUnsafeOps)  return NULL;
  }

577
  return new LibraryIntrinsic(m, is_virtual, is_predicted, does_virtual_dispatch, (vmIntrinsics::ID) id);
D
duke 已提交
578 579 580 581 582 583 584 585
}

//----------------------register_library_intrinsics-----------------------
// Initialize this file's data structures, for each Compile instance.
void Compile::register_library_intrinsics() {
  // Nothing to do here.
}

586
JVMState* LibraryIntrinsic::generate(JVMState* jvms, Parse* parent_parser) {
D
duke 已提交
587 588 589 590
  LibraryCallKit kit(jvms, this);
  Compile* C = kit.C;
  int nodes = C->unique();
#ifndef PRODUCT
591
  if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
D
duke 已提交
592 593 594 595 596
    char buf[1000];
    const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
    tty->print_cr("Intrinsic %s", str);
  }
#endif
597 598
  ciMethod* callee = kit.callee();
  const int bci    = kit.bci();
599

600
  // Try to inline the intrinsic.
D
duke 已提交
601
  if (kit.try_to_inline()) {
602
    if (C->print_intrinsics() || C->print_inlining()) {
603
      C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
D
duke 已提交
604 605 606 607 608 609 610 611
    }
    C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
    if (C->log()) {
      C->log()->elem("intrinsic id='%s'%s nodes='%d'",
                     vmIntrinsics::name_at(intrinsic_id()),
                     (is_virtual() ? " virtual='1'" : ""),
                     C->unique() - nodes);
    }
612 613
    // Push the result from the inlined method onto the stack.
    kit.push_result();
D
duke 已提交
614 615 616
    return kit.transfer_exceptions_into_jvms();
  }

617
  // The intrinsic bailed out
618
  if (C->print_intrinsics() || C->print_inlining()) {
619 620
    if (jvms->has_method()) {
      // Not a root compile.
621
      const char* msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
622
      C->print_inlining(callee, jvms->depth() - 1, bci, msg);
623 624 625
    } else {
      // Root compile
      tty->print("Did not generate intrinsic %s%s at bci:%d in",
626
               vmIntrinsics::name_at(intrinsic_id()),
627
               (is_virtual() ? " (virtual)" : ""), bci);
628
    }
D
duke 已提交
629 630 631 632 633
  }
  C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
  return NULL;
}

634 635 636 637 638 639
Node* LibraryIntrinsic::generate_predicate(JVMState* jvms) {
  LibraryCallKit kit(jvms, this);
  Compile* C = kit.C;
  int nodes = C->unique();
#ifndef PRODUCT
  assert(is_predicted(), "sanity");
640
  if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
641 642 643 644 645
    char buf[1000];
    const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
    tty->print_cr("Predicate for intrinsic %s", str);
  }
#endif
646 647
  ciMethod* callee = kit.callee();
  const int bci    = kit.bci();
648 649 650

  Node* slow_ctl = kit.try_to_predicate();
  if (!kit.failing()) {
651
    if (C->print_intrinsics() || C->print_inlining()) {
652
      C->print_inlining(callee, jvms->depth() - 1, bci, is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
653 654
    }
    C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
655 656 657 658 659 660 661 662 663 664
    if (C->log()) {
      C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
                     vmIntrinsics::name_at(intrinsic_id()),
                     (is_virtual() ? " virtual='1'" : ""),
                     C->unique() - nodes);
    }
    return slow_ctl; // Could be NULL if the check folds.
  }

  // The intrinsic bailed out
665
  if (C->print_intrinsics() || C->print_inlining()) {
666 667 668
    if (jvms->has_method()) {
      // Not a root compile.
      const char* msg = "failed to generate predicate for intrinsic";
669
      C->print_inlining(kit.callee(), jvms->depth() - 1, bci, msg);
670 671
    } else {
      // Root compile
672 673 674
      C->print_inlining_stream()->print("Did not generate predicate for intrinsic %s%s at bci:%d in",
                                        vmIntrinsics::name_at(intrinsic_id()),
                                        (is_virtual() ? " (virtual)" : ""), bci);
675 676 677 678 679 680
    }
  }
  C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
  return NULL;
}

D
duke 已提交
681 682 683 684 685
bool LibraryCallKit::try_to_inline() {
  // Handle symbolic names for otherwise undistinguished boolean switches:
  const bool is_store       = true;
  const bool is_native_ptr  = true;
  const bool is_static      = true;
686
  const bool is_volatile    = true;
D
duke 已提交
687

688 689 690 691 692 693 694 695
  if (!jvms()->has_method()) {
    // Root JVMState has a null method.
    assert(map()->memory()->Opcode() == Op_Parm, "");
    // Insert the memory aliasing node
    set_all_memory(reset_memory());
  }
  assert(merged_memory(), "");

696

D
duke 已提交
697
  switch (intrinsic_id()) {
698 699 700
  case vmIntrinsics::_hashCode:                 return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
  case vmIntrinsics::_identityHashCode:         return inline_native_hashcode(/*!virtual*/ false,         is_static);
  case vmIntrinsics::_getClass:                 return inline_native_getClass();
D
duke 已提交
701 702 703 704 705 706 707 708 709 710

  case vmIntrinsics::_dsin:
  case vmIntrinsics::_dcos:
  case vmIntrinsics::_dtan:
  case vmIntrinsics::_dabs:
  case vmIntrinsics::_datan2:
  case vmIntrinsics::_dsqrt:
  case vmIntrinsics::_dexp:
  case vmIntrinsics::_dlog:
  case vmIntrinsics::_dlog10:
711
  case vmIntrinsics::_dpow:                     return inline_math_native(intrinsic_id());
D
duke 已提交
712 713

  case vmIntrinsics::_min:
714 715
  case vmIntrinsics::_max:                      return inline_min_max(intrinsic_id());

716 717 718 719 720 721 722 723 724 725 726 727
  case vmIntrinsics::_addExactI:                return inline_math_addExactI(false /* add */);
  case vmIntrinsics::_addExactL:                return inline_math_addExactL(false /* add */);
  case vmIntrinsics::_decrementExactI:          return inline_math_subtractExactI(true /* decrement */);
  case vmIntrinsics::_decrementExactL:          return inline_math_subtractExactL(true /* decrement */);
  case vmIntrinsics::_incrementExactI:          return inline_math_addExactI(true /* increment */);
  case vmIntrinsics::_incrementExactL:          return inline_math_addExactL(true /* increment */);
  case vmIntrinsics::_multiplyExactI:           return inline_math_multiplyExactI();
  case vmIntrinsics::_multiplyExactL:           return inline_math_multiplyExactL();
  case vmIntrinsics::_negateExactI:             return inline_math_negateExactI();
  case vmIntrinsics::_negateExactL:             return inline_math_negateExactL();
  case vmIntrinsics::_subtractExactI:           return inline_math_subtractExactI(false /* subtract */);
  case vmIntrinsics::_subtractExactL:           return inline_math_subtractExactL(false /* subtract */);
728

729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811
  case vmIntrinsics::_arraycopy:                return inline_arraycopy();

  case vmIntrinsics::_compareTo:                return inline_string_compareTo();
  case vmIntrinsics::_indexOf:                  return inline_string_indexOf();
  case vmIntrinsics::_equals:                   return inline_string_equals();

  case vmIntrinsics::_getObject:                return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT,  !is_volatile);
  case vmIntrinsics::_getBoolean:               return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN, !is_volatile);
  case vmIntrinsics::_getByte:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE,    !is_volatile);
  case vmIntrinsics::_getShort:                 return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,   !is_volatile);
  case vmIntrinsics::_getChar:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,    !is_volatile);
  case vmIntrinsics::_getInt:                   return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,     !is_volatile);
  case vmIntrinsics::_getLong:                  return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,    !is_volatile);
  case vmIntrinsics::_getFloat:                 return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT,   !is_volatile);
  case vmIntrinsics::_getDouble:                return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE,  !is_volatile);

  case vmIntrinsics::_putObject:                return inline_unsafe_access(!is_native_ptr,  is_store, T_OBJECT,  !is_volatile);
  case vmIntrinsics::_putBoolean:               return inline_unsafe_access(!is_native_ptr,  is_store, T_BOOLEAN, !is_volatile);
  case vmIntrinsics::_putByte:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_BYTE,    !is_volatile);
  case vmIntrinsics::_putShort:                 return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,   !is_volatile);
  case vmIntrinsics::_putChar:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,    !is_volatile);
  case vmIntrinsics::_putInt:                   return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,     !is_volatile);
  case vmIntrinsics::_putLong:                  return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,    !is_volatile);
  case vmIntrinsics::_putFloat:                 return inline_unsafe_access(!is_native_ptr,  is_store, T_FLOAT,   !is_volatile);
  case vmIntrinsics::_putDouble:                return inline_unsafe_access(!is_native_ptr,  is_store, T_DOUBLE,  !is_volatile);

  case vmIntrinsics::_getByte_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_BYTE,    !is_volatile);
  case vmIntrinsics::_getShort_raw:             return inline_unsafe_access( is_native_ptr, !is_store, T_SHORT,   !is_volatile);
  case vmIntrinsics::_getChar_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_CHAR,    !is_volatile);
  case vmIntrinsics::_getInt_raw:               return inline_unsafe_access( is_native_ptr, !is_store, T_INT,     !is_volatile);
  case vmIntrinsics::_getLong_raw:              return inline_unsafe_access( is_native_ptr, !is_store, T_LONG,    !is_volatile);
  case vmIntrinsics::_getFloat_raw:             return inline_unsafe_access( is_native_ptr, !is_store, T_FLOAT,   !is_volatile);
  case vmIntrinsics::_getDouble_raw:            return inline_unsafe_access( is_native_ptr, !is_store, T_DOUBLE,  !is_volatile);
  case vmIntrinsics::_getAddress_raw:           return inline_unsafe_access( is_native_ptr, !is_store, T_ADDRESS, !is_volatile);

  case vmIntrinsics::_putByte_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_BYTE,    !is_volatile);
  case vmIntrinsics::_putShort_raw:             return inline_unsafe_access( is_native_ptr,  is_store, T_SHORT,   !is_volatile);
  case vmIntrinsics::_putChar_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_CHAR,    !is_volatile);
  case vmIntrinsics::_putInt_raw:               return inline_unsafe_access( is_native_ptr,  is_store, T_INT,     !is_volatile);
  case vmIntrinsics::_putLong_raw:              return inline_unsafe_access( is_native_ptr,  is_store, T_LONG,    !is_volatile);
  case vmIntrinsics::_putFloat_raw:             return inline_unsafe_access( is_native_ptr,  is_store, T_FLOAT,   !is_volatile);
  case vmIntrinsics::_putDouble_raw:            return inline_unsafe_access( is_native_ptr,  is_store, T_DOUBLE,  !is_volatile);
  case vmIntrinsics::_putAddress_raw:           return inline_unsafe_access( is_native_ptr,  is_store, T_ADDRESS, !is_volatile);

  case vmIntrinsics::_getObjectVolatile:        return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT,   is_volatile);
  case vmIntrinsics::_getBooleanVolatile:       return inline_unsafe_access(!is_native_ptr, !is_store, T_BOOLEAN,  is_volatile);
  case vmIntrinsics::_getByteVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_BYTE,     is_volatile);
  case vmIntrinsics::_getShortVolatile:         return inline_unsafe_access(!is_native_ptr, !is_store, T_SHORT,    is_volatile);
  case vmIntrinsics::_getCharVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_CHAR,     is_volatile);
  case vmIntrinsics::_getIntVolatile:           return inline_unsafe_access(!is_native_ptr, !is_store, T_INT,      is_volatile);
  case vmIntrinsics::_getLongVolatile:          return inline_unsafe_access(!is_native_ptr, !is_store, T_LONG,     is_volatile);
  case vmIntrinsics::_getFloatVolatile:         return inline_unsafe_access(!is_native_ptr, !is_store, T_FLOAT,    is_volatile);
  case vmIntrinsics::_getDoubleVolatile:        return inline_unsafe_access(!is_native_ptr, !is_store, T_DOUBLE,   is_volatile);

  case vmIntrinsics::_putObjectVolatile:        return inline_unsafe_access(!is_native_ptr,  is_store, T_OBJECT,   is_volatile);
  case vmIntrinsics::_putBooleanVolatile:       return inline_unsafe_access(!is_native_ptr,  is_store, T_BOOLEAN,  is_volatile);
  case vmIntrinsics::_putByteVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_BYTE,     is_volatile);
  case vmIntrinsics::_putShortVolatile:         return inline_unsafe_access(!is_native_ptr,  is_store, T_SHORT,    is_volatile);
  case vmIntrinsics::_putCharVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_CHAR,     is_volatile);
  case vmIntrinsics::_putIntVolatile:           return inline_unsafe_access(!is_native_ptr,  is_store, T_INT,      is_volatile);
  case vmIntrinsics::_putLongVolatile:          return inline_unsafe_access(!is_native_ptr,  is_store, T_LONG,     is_volatile);
  case vmIntrinsics::_putFloatVolatile:         return inline_unsafe_access(!is_native_ptr,  is_store, T_FLOAT,    is_volatile);
  case vmIntrinsics::_putDoubleVolatile:        return inline_unsafe_access(!is_native_ptr,  is_store, T_DOUBLE,   is_volatile);

  case vmIntrinsics::_prefetchRead:             return inline_unsafe_prefetch(!is_native_ptr, !is_store, !is_static);
  case vmIntrinsics::_prefetchWrite:            return inline_unsafe_prefetch(!is_native_ptr,  is_store, !is_static);
  case vmIntrinsics::_prefetchReadStatic:       return inline_unsafe_prefetch(!is_native_ptr, !is_store,  is_static);
  case vmIntrinsics::_prefetchWriteStatic:      return inline_unsafe_prefetch(!is_native_ptr,  is_store,  is_static);

  case vmIntrinsics::_compareAndSwapObject:     return inline_unsafe_load_store(T_OBJECT, LS_cmpxchg);
  case vmIntrinsics::_compareAndSwapInt:        return inline_unsafe_load_store(T_INT,    LS_cmpxchg);
  case vmIntrinsics::_compareAndSwapLong:       return inline_unsafe_load_store(T_LONG,   LS_cmpxchg);

  case vmIntrinsics::_putOrderedObject:         return inline_unsafe_ordered_store(T_OBJECT);
  case vmIntrinsics::_putOrderedInt:            return inline_unsafe_ordered_store(T_INT);
  case vmIntrinsics::_putOrderedLong:           return inline_unsafe_ordered_store(T_LONG);

  case vmIntrinsics::_getAndAddInt:             return inline_unsafe_load_store(T_INT,    LS_xadd);
  case vmIntrinsics::_getAndAddLong:            return inline_unsafe_load_store(T_LONG,   LS_xadd);
  case vmIntrinsics::_getAndSetInt:             return inline_unsafe_load_store(T_INT,    LS_xchg);
  case vmIntrinsics::_getAndSetLong:            return inline_unsafe_load_store(T_LONG,   LS_xchg);
  case vmIntrinsics::_getAndSetObject:          return inline_unsafe_load_store(T_OBJECT, LS_xchg);

812 813 814 815
  case vmIntrinsics::_loadFence:
  case vmIntrinsics::_storeFence:
  case vmIntrinsics::_fullFence:                return inline_unsafe_fence(intrinsic_id());

816 817
  case vmIntrinsics::_currentThread:            return inline_native_currentThread();
  case vmIntrinsics::_isInterrupted:            return inline_native_isInterrupted();
D
duke 已提交
818

819
#ifdef TRACE_HAVE_INTRINSICS
820 821 822
  case vmIntrinsics::_classID:                  return inline_native_classID();
  case vmIntrinsics::_threadID:                 return inline_native_threadID();
  case vmIntrinsics::_counterTime:              return inline_native_time_funcs(CAST_FROM_FN_PTR(address, TRACE_TIME_METHOD), "counterTime");
823
#endif
824 825 826 827 828 829 830 831 832 833 834 835
  case vmIntrinsics::_currentTimeMillis:        return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
  case vmIntrinsics::_nanoTime:                 return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
  case vmIntrinsics::_allocateInstance:         return inline_unsafe_allocate();
  case vmIntrinsics::_copyMemory:               return inline_unsafe_copyMemory();
  case vmIntrinsics::_newArray:                 return inline_native_newArray();
  case vmIntrinsics::_getLength:                return inline_native_getLength();
  case vmIntrinsics::_copyOf:                   return inline_array_copyOf(false);
  case vmIntrinsics::_copyOfRange:              return inline_array_copyOf(true);
  case vmIntrinsics::_equalsC:                  return inline_array_equals();
  case vmIntrinsics::_clone:                    return inline_native_clone(intrinsic()->is_virtual());

  case vmIntrinsics::_isAssignableFrom:         return inline_native_subtype_check();
D
duke 已提交
836 837 838 839 840 841 842 843

  case vmIntrinsics::_isInstance:
  case vmIntrinsics::_getModifiers:
  case vmIntrinsics::_isInterface:
  case vmIntrinsics::_isArray:
  case vmIntrinsics::_isPrimitive:
  case vmIntrinsics::_getSuperclass:
  case vmIntrinsics::_getComponentType:
844
  case vmIntrinsics::_getClassAccessFlags:      return inline_native_Class_query(intrinsic_id());
D
duke 已提交
845 846 847 848 849 850

  case vmIntrinsics::_floatToRawIntBits:
  case vmIntrinsics::_floatToIntBits:
  case vmIntrinsics::_intBitsToFloat:
  case vmIntrinsics::_doubleToRawLongBits:
  case vmIntrinsics::_doubleToLongBits:
851
  case vmIntrinsics::_longBitsToDouble:         return inline_fp_conversions(intrinsic_id());
D
duke 已提交
852

853 854 855 856
  case vmIntrinsics::_numberOfLeadingZeros_i:
  case vmIntrinsics::_numberOfLeadingZeros_l:
  case vmIntrinsics::_numberOfTrailingZeros_i:
  case vmIntrinsics::_numberOfTrailingZeros_l:
857 858
  case vmIntrinsics::_bitCount_i:
  case vmIntrinsics::_bitCount_l:
D
duke 已提交
859 860
  case vmIntrinsics::_reverseBytes_i:
  case vmIntrinsics::_reverseBytes_l:
861
  case vmIntrinsics::_reverseBytes_s:
862
  case vmIntrinsics::_reverseBytes_c:           return inline_number_methods(intrinsic_id());
D
duke 已提交
863

864
  case vmIntrinsics::_getCallerClass:           return inline_native_Reflection_getCallerClass();
D
duke 已提交
865

866
  case vmIntrinsics::_Reference_get:            return inline_reference_get();
867

868
  case vmIntrinsics::_aescrypt_encryptBlock:
869
  case vmIntrinsics::_aescrypt_decryptBlock:    return inline_aescrypt_Block(intrinsic_id());
870 871 872 873 874

  case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
  case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
    return inline_cipherBlockChaining_AESCrypt(intrinsic_id());

875 876 877
  case vmIntrinsics::_encodeISOArray:
    return inline_encodeISOArray();

878 879 880 881 882 883 884
  case vmIntrinsics::_updateCRC32:
    return inline_updateCRC32();
  case vmIntrinsics::_updateBytesCRC32:
    return inline_updateBytesCRC32();
  case vmIntrinsics::_updateByteBufferCRC32:
    return inline_updateByteBufferCRC32();

D
duke 已提交
885 886 887 888 889 890 891 892 893 894 895 896 897
  default:
    // If you get here, it may be that someone has added a new intrinsic
    // to the list in vmSymbols.hpp without implementing it here.
#ifndef PRODUCT
    if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
      tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
                    vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
    }
#endif
    return false;
  }
}

898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927
Node* LibraryCallKit::try_to_predicate() {
  if (!jvms()->has_method()) {
    // Root JVMState has a null method.
    assert(map()->memory()->Opcode() == Op_Parm, "");
    // Insert the memory aliasing node
    set_all_memory(reset_memory());
  }
  assert(merged_memory(), "");

  switch (intrinsic_id()) {
  case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
    return inline_cipherBlockChaining_AESCrypt_predicate(false);
  case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
    return inline_cipherBlockChaining_AESCrypt_predicate(true);

  default:
    // If you get here, it may be that someone has added a new intrinsic
    // to the list in vmSymbols.hpp without implementing it here.
#ifndef PRODUCT
    if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
      tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
                    vmIntrinsics::name_at(intrinsic_id()), intrinsic_id());
    }
#endif
    Node* slow_ctl = control();
    set_control(top()); // No fast path instrinsic
    return slow_ctl;
  }
}

928
//------------------------------set_result-------------------------------
D
duke 已提交
929
// Helper function for finishing intrinsics.
930
void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
D
duke 已提交
931 932
  record_for_igvn(region);
  set_control(_gvn.transform(region));
933 934
  set_result( _gvn.transform(value));
  assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
D
duke 已提交
935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961
}

//------------------------------generate_guard---------------------------
// Helper function for generating guarded fast-slow graph structures.
// The given 'test', if true, guards a slow path.  If the test fails
// then a fast path can be taken.  (We generally hope it fails.)
// In all cases, GraphKit::control() is updated to the fast path.
// The returned value represents the control for the slow path.
// The return value is never 'top'; it is either a valid control
// or NULL if it is obvious that the slow path can never be taken.
// Also, if region and the slow control are not NULL, the slow edge
// is appended to the region.
Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
  if (stopped()) {
    // Already short circuited.
    return NULL;
  }

  // Build an if node and its projections.
  // If test is true we take the slow path, which we assume is uncommon.
  if (_gvn.type(test) == TypeInt::ZERO) {
    // The slow branch is never taken.  No need to build this guard.
    return NULL;
  }

  IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);

962
  Node* if_slow = _gvn.transform(new (C) IfTrueNode(iff));
D
duke 已提交
963 964 965 966 967 968 969 970
  if (if_slow == top()) {
    // The slow branch is never taken.  No need to build this guard.
    return NULL;
  }

  if (region != NULL)
    region->add_req(if_slow);

971
  Node* if_fast = _gvn.transform(new (C) IfFalseNode(iff));
D
duke 已提交
972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989
  set_control(if_fast);

  return if_slow;
}

inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
  return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
}
inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
  return generate_guard(test, region, PROB_FAIR);
}

inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
                                                     Node* *pos_index) {
  if (stopped())
    return NULL;                // already stopped
  if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
    return NULL;                // index is already adequately typed
990 991
  Node* cmp_lt = _gvn.transform(new (C) CmpINode(index, intcon(0)));
  Node* bol_lt = _gvn.transform(new (C) BoolNode(cmp_lt, BoolTest::lt));
D
duke 已提交
992 993 994
  Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
  if (is_neg != NULL && pos_index != NULL) {
    // Emulate effect of Parse::adjust_map_after_if.
995
    Node* ccast = new (C) CastIINode(index, TypeInt::POS);
D
duke 已提交
996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007
    ccast->set_req(0, control());
    (*pos_index) = _gvn.transform(ccast);
  }
  return is_neg;
}

inline Node* LibraryCallKit::generate_nonpositive_guard(Node* index, bool never_negative,
                                                        Node* *pos_index) {
  if (stopped())
    return NULL;                // already stopped
  if (_gvn.type(index)->higher_equal(TypeInt::POS1)) // [1,maxint]
    return NULL;                // index is already adequately typed
1008
  Node* cmp_le = _gvn.transform(new (C) CmpINode(index, intcon(0)));
D
duke 已提交
1009
  BoolTest::mask le_or_eq = (never_negative ? BoolTest::eq : BoolTest::le);
1010
  Node* bol_le = _gvn.transform(new (C) BoolNode(cmp_le, le_or_eq));
D
duke 已提交
1011 1012 1013
  Node* is_notp = generate_guard(bol_le, NULL, PROB_MIN);
  if (is_notp != NULL && pos_index != NULL) {
    // Emulate effect of Parse::adjust_map_after_if.
1014
    Node* ccast = new (C) CastIINode(index, TypeInt::POS1);
D
duke 已提交
1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041
    ccast->set_req(0, control());
    (*pos_index) = _gvn.transform(ccast);
  }
  return is_notp;
}

// Make sure that 'position' is a valid limit index, in [0..length].
// There are two equivalent plans for checking this:
//   A. (offset + copyLength)  unsigned<=  arrayLength
//   B. offset  <=  (arrayLength - copyLength)
// We require that all of the values above, except for the sum and
// difference, are already known to be non-negative.
// Plan A is robust in the face of overflow, if offset and copyLength
// are both hugely positive.
//
// Plan B is less direct and intuitive, but it does not overflow at
// all, since the difference of two non-negatives is always
// representable.  Whenever Java methods must perform the equivalent
// check they generally use Plan B instead of Plan A.
// For the moment we use Plan A.
inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
                                                  Node* subseq_length,
                                                  Node* array_length,
                                                  RegionNode* region) {
  if (stopped())
    return NULL;                // already stopped
  bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
K
kvn 已提交
1042
  if (zero_offset && subseq_length->eqv_uncast(array_length))
D
duke 已提交
1043 1044 1045
    return NULL;                // common case of whole-array copy
  Node* last = subseq_length;
  if (!zero_offset)             // last += offset
1046 1047 1048
    last = _gvn.transform(new (C) AddINode(last, offset));
  Node* cmp_lt = _gvn.transform(new (C) CmpUNode(array_length, last));
  Node* bol_lt = _gvn.transform(new (C) BoolNode(cmp_lt, BoolTest::lt));
D
duke 已提交
1049 1050 1051 1052 1053 1054 1055 1056 1057
  Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
  return is_over;
}


//--------------------------generate_current_thread--------------------
Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
  ciKlass*    thread_klass = env()->Thread_klass();
  const Type* thread_type  = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1058
  Node* thread = _gvn.transform(new (C) ThreadLocalNode());
D
duke 已提交
1059 1060 1061 1062 1063 1064 1065
  Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::threadObj_offset()));
  Node* threadObj = make_load(NULL, p, thread_type, T_OBJECT);
  tls_output = thread;
  return threadObj;
}


1066
//------------------------------make_string_method_node------------------------
K
kvn 已提交
1067 1068 1069 1070
// Helper method for String intrinsic functions. This version is called
// with str1 and str2 pointing to String object nodes.
//
Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1, Node* str2) {
1071 1072
  Node* no_ctrl = NULL;

K
kvn 已提交
1073 1074 1075
  // Get start addr of string
  Node* str1_value   = load_String_value(no_ctrl, str1);
  Node* str1_offset  = load_String_offset(no_ctrl, str1);
1076 1077
  Node* str1_start   = array_element_address(str1_value, str1_offset, T_CHAR);

K
kvn 已提交
1078 1079 1080 1081 1082
  // Get length of string 1
  Node* str1_len  = load_String_length(no_ctrl, str1);

  Node* str2_value   = load_String_value(no_ctrl, str2);
  Node* str2_offset  = load_String_offset(no_ctrl, str2);
1083 1084
  Node* str2_start   = array_element_address(str2_value, str2_offset, T_CHAR);

K
kvn 已提交
1085 1086 1087 1088 1089 1090 1091 1092
  Node* str2_len = NULL;
  Node* result = NULL;

  switch (opcode) {
  case Op_StrIndexOf:
    // Get length of string 2
    str2_len = load_String_length(no_ctrl, str2);

1093
    result = new (C) StrIndexOfNode(control(), memory(TypeAryPtr::CHARS),
K
kvn 已提交
1094 1095 1096 1097 1098 1099
                                 str1_start, str1_len, str2_start, str2_len);
    break;
  case Op_StrComp:
    // Get length of string 2
    str2_len = load_String_length(no_ctrl, str2);

1100
    result = new (C) StrCompNode(control(), memory(TypeAryPtr::CHARS),
K
kvn 已提交
1101 1102 1103
                                 str1_start, str1_len, str2_start, str2_len);
    break;
  case Op_StrEquals:
1104
    result = new (C) StrEqualsNode(control(), memory(TypeAryPtr::CHARS),
K
kvn 已提交
1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
                               str1_start, str2_start, str1_len);
    break;
  default:
    ShouldNotReachHere();
    return NULL;
  }

  // All these intrinsics have checks.
  C->set_has_split_ifs(true); // Has chance for split-if optimization

  return _gvn.transform(result);
}

// Helper method for String intrinsic functions. This version is called
// with str1 and str2 pointing to char[] nodes, with cnt1 and cnt2 pointing
// to Int nodes containing the lenghts of str1 and str2.
//
Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2) {
1123 1124 1125
  Node* result = NULL;
  switch (opcode) {
  case Op_StrIndexOf:
1126
    result = new (C) StrIndexOfNode(control(), memory(TypeAryPtr::CHARS),
K
kvn 已提交
1127
                                 str1_start, cnt1, str2_start, cnt2);
1128 1129
    break;
  case Op_StrComp:
1130
    result = new (C) StrCompNode(control(), memory(TypeAryPtr::CHARS),
K
kvn 已提交
1131
                                 str1_start, cnt1, str2_start, cnt2);
1132 1133
    break;
  case Op_StrEquals:
1134
    result = new (C) StrEqualsNode(control(), memory(TypeAryPtr::CHARS),
K
kvn 已提交
1135
                                 str1_start, str2_start, cnt1);
1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147
    break;
  default:
    ShouldNotReachHere();
    return NULL;
  }

  // All these intrinsics have checks.
  C->set_has_split_ifs(true); // Has chance for split-if optimization

  return _gvn.transform(result);
}

D
duke 已提交
1148
//------------------------------inline_string_compareTo------------------------
1149
// public int java.lang.String.compareTo(String anotherString);
D
duke 已提交
1150
bool LibraryCallKit::inline_string_compareTo() {
1151 1152
  Node* receiver = null_check(argument(0));
  Node* arg      = null_check(argument(1));
D
duke 已提交
1153 1154 1155
  if (stopped()) {
    return true;
  }
1156
  set_result(make_string_method_node(Op_StrComp, receiver, arg));
D
duke 已提交
1157 1158 1159
  return true;
}

C
cfang 已提交
1160 1161
//------------------------------inline_string_equals------------------------
bool LibraryCallKit::inline_string_equals() {
1162 1163 1164 1165
  Node* receiver = null_check_receiver();
  // NOTE: Do not null check argument for String.equals() because spec
  // allows to specify NULL as argument.
  Node* argument = this->argument(1);
C
cfang 已提交
1166 1167 1168 1169
  if (stopped()) {
    return true;
  }

1170
  // paths (plus control) merge
1171 1172
  RegionNode* region = new (C) RegionNode(5);
  Node* phi = new (C) PhiNode(region, TypeInt::BOOL);
1173 1174

  // does source == target string?
1175 1176
  Node* cmp = _gvn.transform(new (C) CmpPNode(receiver, argument));
  Node* bol = _gvn.transform(new (C) BoolNode(cmp, BoolTest::eq));
1177 1178 1179 1180 1181 1182 1183 1184

  Node* if_eq = generate_slow_guard(bol, NULL);
  if (if_eq != NULL) {
    // receiver == argument
    phi->init_req(2, intcon(1));
    region->init_req(2, if_eq);
  }

C
cfang 已提交
1185 1186 1187
  // get String klass for instanceOf
  ciInstanceKlass* klass = env()->String_klass();

1188 1189
  if (!stopped()) {
    Node* inst = gen_instanceof(argument, makecon(TypeKlassPtr::make(klass)));
1190 1191
    Node* cmp  = _gvn.transform(new (C) CmpINode(inst, intcon(1)));
    Node* bol  = _gvn.transform(new (C) BoolNode(cmp, BoolTest::ne));
C
cfang 已提交
1192

1193 1194
    Node* inst_false = generate_guard(bol, NULL, PROB_MIN);
    //instanceOf == true, fallthrough
C
cfang 已提交
1195

1196 1197 1198 1199 1200
    if (inst_false != NULL) {
      phi->init_req(3, intcon(0));
      region->init_req(3, inst_false);
    }
  }
C
cfang 已提交
1201

1202
  if (!stopped()) {
K
kvn 已提交
1203 1204
    const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(klass);

1205
    // Properly cast the argument to String
1206
    argument = _gvn.transform(new (C) CheckCastPPNode(control(), argument, string_type));
1207 1208
    // This path is taken only when argument's type is String:NotNull.
    argument = cast_not_null(argument, false);
1209

K
kvn 已提交
1210 1211 1212 1213 1214 1215 1216 1217 1218
    Node* no_ctrl = NULL;

    // Get start addr of receiver
    Node* receiver_val    = load_String_value(no_ctrl, receiver);
    Node* receiver_offset = load_String_offset(no_ctrl, receiver);
    Node* receiver_start = array_element_address(receiver_val, receiver_offset, T_CHAR);

    // Get length of receiver
    Node* receiver_cnt  = load_String_length(no_ctrl, receiver);
1219

K
kvn 已提交
1220
    // Get start addr of argument
1221
    Node* argument_val    = load_String_value(no_ctrl, argument);
K
kvn 已提交
1222 1223 1224 1225 1226
    Node* argument_offset = load_String_offset(no_ctrl, argument);
    Node* argument_start = array_element_address(argument_val, argument_offset, T_CHAR);

    // Get length of argument
    Node* argument_cnt  = load_String_length(no_ctrl, argument);
1227 1228

    // Check for receiver count != argument count
1229 1230
    Node* cmp = _gvn.transform(new(C) CmpINode(receiver_cnt, argument_cnt));
    Node* bol = _gvn.transform(new(C) BoolNode(cmp, BoolTest::ne));
1231 1232 1233 1234 1235
    Node* if_ne = generate_slow_guard(bol, NULL);
    if (if_ne != NULL) {
      phi->init_req(4, intcon(0));
      region->init_req(4, if_ne);
    }
C
cfang 已提交
1236

K
kvn 已提交
1237
    // Check for count == 0 is done by assembler code for StrEquals.
C
cfang 已提交
1238

K
kvn 已提交
1239 1240 1241 1242 1243
    if (!stopped()) {
      Node* equals = make_string_method_node(Op_StrEquals, receiver_start, receiver_cnt, argument_start, argument_cnt);
      phi->init_req(1, equals);
      region->init_req(1, control());
    }
1244
  }
C
cfang 已提交
1245 1246 1247 1248 1249

  // post merge
  set_control(_gvn.transform(region));
  record_for_igvn(region);

1250
  set_result(_gvn.transform(phi));
C
cfang 已提交
1251 1252 1253
  return true;
}

1254 1255
//------------------------------inline_array_equals----------------------------
bool LibraryCallKit::inline_array_equals() {
1256 1257 1258
  Node* arg1 = argument(0);
  Node* arg2 = argument(1);
  set_result(_gvn.transform(new (C) AryEqNode(control(), memory(TypeAryPtr::CHARS), arg1, arg2)));
1259 1260 1261
  return true;
}

D
duke 已提交
1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
// Java version of String.indexOf(constant string)
// class StringDecl {
//   StringDecl(char[] ca) {
//     offset = 0;
//     count = ca.length;
//     value = ca;
//   }
//   int offset;
//   int count;
//   char[] value;
// }
//
// static int string_indexOf_J(StringDecl string_object, char[] target_object,
//                             int targetOffset, int cache_i, int md2) {
//   int cache = cache_i;
//   int sourceOffset = string_object.offset;
//   int sourceCount = string_object.count;
//   int targetCount = target_object.length;
//
//   int targetCountLess1 = targetCount - 1;
//   int sourceEnd = sourceOffset + sourceCount - targetCountLess1;
//
//   char[] source = string_object.value;
//   char[] target = target_object;
//   int lastChar = target[targetCountLess1];
//
//  outer_loop:
//   for (int i = sourceOffset; i < sourceEnd; ) {
//     int src = source[i + targetCountLess1];
//     if (src == lastChar) {
//       // With random strings and a 4-character alphabet,
//       // reverse matching at this point sets up 0.8% fewer
//       // frames, but (paradoxically) makes 0.3% more probes.
//       // Since those probes are nearer the lastChar probe,
//       // there is may be a net D$ win with reverse matching.
//       // But, reversing loop inhibits unroll of inner loop
//       // for unknown reason.  So, does running outer loop from
//       // (sourceOffset - targetCountLess1) to (sourceOffset + sourceCount)
//       for (int j = 0; j < targetCountLess1; j++) {
//         if (target[targetOffset + j] != source[i+j]) {
//           if ((cache & (1 << source[i+j])) == 0) {
//             if (md2 < j+1) {
//               i += j+1;
//               continue outer_loop;
//             }
//           }
//           i += md2;
//           continue outer_loop;
//         }
//       }
//       return i - sourceOffset;
//     }
//     if ((cache & (1 << src)) == 0) {
//       i += targetCountLess1;
//     } // using "i += targetCount;" and an "else i++;" causes a jump to jump.
//     i++;
//   }
//   return -1;
// }

//------------------------------string_indexOf------------------------
Node* LibraryCallKit::string_indexOf(Node* string_object, ciTypeArray* target_array, jint targetOffset_i,
                                     jint cache_i, jint md2_i) {

  Node* no_ctrl  = NULL;
  float likely   = PROB_LIKELY(0.9);
  float unlikely = PROB_UNLIKELY(0.9);

1330
  const int nargs = 0; // no arguments to push back for uncommon trap in predicate
1331

K
kvn 已提交
1332 1333 1334
  Node* source        = load_String_value(no_ctrl, string_object);
  Node* sourceOffset  = load_String_offset(no_ctrl, string_object);
  Node* sourceCount   = load_String_length(no_ctrl, string_object);
D
duke 已提交
1335

1336
  Node* target = _gvn.transform( makecon(TypeOopPtr::make_from_constant(target_array, true)));
D
duke 已提交
1337 1338 1339 1340
  jint target_length = target_array->length();
  const TypeAry* target_array_type = TypeAry::make(TypeInt::CHAR, TypeInt::make(0, target_length, Type::WidenMin));
  const TypeAryPtr* target_type = TypeAryPtr::make(TypePtr::BotPTR, target_array_type, target_array->klass(), true, Type::OffsetBot);

1341 1342 1343 1344 1345
  // String.value field is known to be @Stable.
  if (UseImplicitStableValues) {
    target = cast_array_to_stable(target, target_type);
  }

1346
  IdealKit kit(this, false, true);
D
duke 已提交
1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357
#define __ kit.
  Node* zero             = __ ConI(0);
  Node* one              = __ ConI(1);
  Node* cache            = __ ConI(cache_i);
  Node* md2              = __ ConI(md2_i);
  Node* lastChar         = __ ConI(target_array->char_at(target_length - 1));
  Node* targetCount      = __ ConI(target_length);
  Node* targetCountLess1 = __ ConI(target_length - 1);
  Node* targetOffset     = __ ConI(targetOffset_i);
  Node* sourceEnd        = __ SubI(__ AddI(sourceOffset, sourceCount), targetCountLess1);

1358
  IdealVariable rtn(kit), i(kit), j(kit); __ declarations_done();
D
duke 已提交
1359 1360 1361 1362
  Node* outer_loop = __ make_label(2 /* goto */);
  Node* return_    = __ make_label(1);

  __ set(rtn,__ ConI(-1));
1363
  __ loop(this, nargs, i, sourceOffset, BoolTest::lt, sourceEnd); {
D
duke 已提交
1364 1365 1366 1367
       Node* i2  = __ AddI(__ value(i), targetCountLess1);
       // pin to prohibit loading of "next iteration" value which may SEGV (rare)
       Node* src = load_array_element(__ ctrl(), source, i2, TypeAryPtr::CHARS);
       __ if_then(src, BoolTest::eq, lastChar, unlikely); {
1368
         __ loop(this, nargs, j, zero, BoolTest::lt, targetCountLess1); {
D
duke 已提交
1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395
              Node* tpj = __ AddI(targetOffset, __ value(j));
              Node* targ = load_array_element(no_ctrl, target, tpj, target_type);
              Node* ipj  = __ AddI(__ value(i), __ value(j));
              Node* src2 = load_array_element(no_ctrl, source, ipj, TypeAryPtr::CHARS);
              __ if_then(targ, BoolTest::ne, src2); {
                __ if_then(__ AndI(cache, __ LShiftI(one, src2)), BoolTest::eq, zero); {
                  __ if_then(md2, BoolTest::lt, __ AddI(__ value(j), one)); {
                    __ increment(i, __ AddI(__ value(j), one));
                    __ goto_(outer_loop);
                  } __ end_if(); __ dead(j);
                }__ end_if(); __ dead(j);
                __ increment(i, md2);
                __ goto_(outer_loop);
              }__ end_if();
              __ increment(j, one);
         }__ end_loop(); __ dead(j);
         __ set(rtn, __ SubI(__ value(i), sourceOffset)); __ dead(i);
         __ goto_(return_);
       }__ end_if();
       __ if_then(__ AndI(cache, __ LShiftI(one, src)), BoolTest::eq, zero, likely); {
         __ increment(i, targetCountLess1);
       }__ end_if();
       __ increment(i, one);
       __ bind(outer_loop);
  }__ end_loop(); __ dead(i);
  __ bind(return_);

1396
  // Final sync IdealKit and GraphKit.
1397
  final_sync(kit);
D
duke 已提交
1398 1399 1400 1401 1402 1403 1404 1405
  Node* result = __ value(rtn);
#undef __
  C->set_has_loops(true);
  return result;
}

//------------------------------inline_string_indexOf------------------------
bool LibraryCallKit::inline_string_indexOf() {
1406 1407
  Node* receiver = argument(0);
  Node* arg      = argument(1);
D
duke 已提交
1408

C
cfang 已提交
1409
  Node* result;
1410 1411
  // Disable the use of pcmpestri until it can be guaranteed that
  // the load doesn't cross into the uncommited space.
1412
  if (Matcher::has_match_rule(Op_StrIndexOf) &&
C
cfang 已提交
1413 1414 1415 1416
      UseSSE42Intrinsics) {
    // Generate SSE4.2 version of indexOf
    // We currently only have match rules that use SSE4.2

1417 1418
    receiver = null_check(receiver);
    arg      = null_check(arg);
C
cfang 已提交
1419 1420 1421
    if (stopped()) {
      return true;
    }
D
duke 已提交
1422

1423 1424 1425
    ciInstanceKlass* str_klass = env()->String_klass();
    const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(str_klass);

1426
    // Make the merge point
1427 1428
    RegionNode* result_rgn = new (C) RegionNode(4);
    Node*       result_phi = new (C) PhiNode(result_rgn, TypeInt::INT);
1429 1430
    Node* no_ctrl  = NULL;

K
kvn 已提交
1431 1432 1433 1434 1435 1436 1437 1438 1439
    // Get start addr of source string
    Node* source = load_String_value(no_ctrl, receiver);
    Node* source_offset = load_String_offset(no_ctrl, receiver);
    Node* source_start = array_element_address(source, source_offset, T_CHAR);

    // Get length of source string
    Node* source_cnt  = load_String_length(no_ctrl, receiver);

    // Get start addr of substring
1440 1441
    Node* substr = load_String_value(no_ctrl, arg);
    Node* substr_offset = load_String_offset(no_ctrl, arg);
K
kvn 已提交
1442
    Node* substr_start = array_element_address(substr, substr_offset, T_CHAR);
1443

K
kvn 已提交
1444
    // Get length of source string
1445
    Node* substr_cnt  = load_String_length(no_ctrl, arg);
1446 1447

    // Check for substr count > string count
1448 1449
    Node* cmp = _gvn.transform(new(C) CmpINode(substr_cnt, source_cnt));
    Node* bol = _gvn.transform(new(C) BoolNode(cmp, BoolTest::gt));
1450 1451 1452 1453 1454 1455
    Node* if_gt = generate_slow_guard(bol, NULL);
    if (if_gt != NULL) {
      result_phi->init_req(2, intcon(-1));
      result_rgn->init_req(2, if_gt);
    }

1456 1457
    if (!stopped()) {
      // Check for substr count == 0
1458 1459
      cmp = _gvn.transform(new(C) CmpINode(substr_cnt, intcon(0)));
      bol = _gvn.transform(new(C) BoolNode(cmp, BoolTest::eq));
1460 1461 1462 1463 1464 1465 1466
      Node* if_zero = generate_slow_guard(bol, NULL);
      if (if_zero != NULL) {
        result_phi->init_req(3, intcon(0));
        result_rgn->init_req(3, if_zero);
      }
    }

1467
    if (!stopped()) {
K
kvn 已提交
1468
      result = make_string_method_node(Op_StrIndexOf, source_start, source_cnt, substr_start, substr_cnt);
1469 1470 1471 1472 1473 1474 1475
      result_phi->init_req(1, result);
      result_rgn->init_req(1, control());
    }
    set_control(_gvn.transform(result_rgn));
    record_for_igvn(result_rgn);
    result = _gvn.transform(result_phi);

1476 1477
  } else { // Use LibraryCallKit::string_indexOf
    // don't intrinsify if argument isn't a constant string.
1478
    if (!arg->is_Con()) {
C
cfang 已提交
1479 1480
     return false;
    }
1481
    const TypeOopPtr* str_type = _gvn.type(arg)->isa_oopptr();
C
cfang 已提交
1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492
    if (str_type == NULL) {
      return false;
    }
    ciInstanceKlass* klass = env()->String_klass();
    ciObject* str_const = str_type->const_oop();
    if (str_const == NULL || str_const->klass() != klass) {
      return false;
    }
    ciInstance* str = str_const->as_instance();
    assert(str != NULL, "must be instance");

K
kvn 已提交
1493
    ciObject* v = str->field_value_by_offset(java_lang_String::value_offset_in_bytes()).as_object();
C
cfang 已提交
1494 1495
    ciTypeArray* pat = v->as_type_array(); // pattern (argument) character array

K
kvn 已提交
1496 1497 1498 1499 1500 1501 1502 1503 1504 1505
    int o;
    int c;
    if (java_lang_String::has_offset_field()) {
      o = str->field_value_by_offset(java_lang_String::offset_offset_in_bytes()).as_int();
      c = str->field_value_by_offset(java_lang_String::count_offset_in_bytes()).as_int();
    } else {
      o = 0;
      c = pat->length();
    }

C
cfang 已提交
1506 1507 1508 1509 1510
    // constant strings have no offset and count == length which
    // simplifies the resulting code somewhat so lets optimize for that.
    if (o != 0 || c != pat->length()) {
     return false;
    }
D
duke 已提交
1511

1512 1513
    receiver = null_check(receiver, T_OBJECT);
    // NOTE: No null check on the argument is needed since it's a constant String oop.
C
cfang 已提交
1514
    if (stopped()) {
1515
      return true;
C
cfang 已提交
1516
    }
D
duke 已提交
1517

C
cfang 已提交
1518 1519
    // The null string as a pattern always returns 0 (match at beginning of string)
    if (c == 0) {
1520
      set_result(intcon(0));
C
cfang 已提交
1521 1522
      return true;
    }
D
duke 已提交
1523

C
cfang 已提交
1524 1525 1526 1527 1528 1529 1530
    // Generate default indexOf
    jchar lastChar = pat->char_at(o + (c - 1));
    int cache = 0;
    int i;
    for (i = 0; i < c - 1; i++) {
      assert(i < pat->length(), "out of range");
      cache |= (1 << (pat->char_at(o + i) & (sizeof(cache) * BitsPerByte - 1)));
D
duke 已提交
1531
    }
C
cfang 已提交
1532 1533 1534 1535 1536 1537 1538 1539 1540 1541

    int md2 = c;
    for (i = 0; i < c - 1; i++) {
      assert(i < pat->length(), "out of range");
      if (pat->char_at(o + i) == lastChar) {
        md2 = (c - 1) - i;
      }
    }

    result = string_indexOf(receiver, pat, o, cache, md2);
D
duke 已提交
1542
  }
1543
  set_result(result);
D
duke 已提交
1544 1545 1546
  return true;
}

1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563
//--------------------------round_double_node--------------------------------
// Round a double node if necessary.
Node* LibraryCallKit::round_double_node(Node* n) {
  if (Matcher::strict_fp_requires_explicit_rounding && UseSSE <= 1)
    n = _gvn.transform(new (C) RoundDoubleNode(0, n));
  return n;
}

//------------------------------inline_math-----------------------------------
// public static double Math.abs(double)
// public static double Math.sqrt(double)
// public static double Math.log(double)
// public static double Math.log10(double)
bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
  Node* arg = round_double_node(argument(0));
  Node* n;
  switch (id) {
1564 1565 1566 1567
  case vmIntrinsics::_dabs:   n = new (C) AbsDNode(                arg);  break;
  case vmIntrinsics::_dsqrt:  n = new (C) SqrtDNode(C, control(),  arg);  break;
  case vmIntrinsics::_dlog:   n = new (C) LogDNode(C, control(),   arg);  break;
  case vmIntrinsics::_dlog10: n = new (C) Log10DNode(C, control(), arg);  break;
1568 1569 1570 1571
  default:  fatal_unexpected_iid(id);  break;
  }
  set_result(_gvn.transform(n));
  return true;
D
duke 已提交
1572 1573 1574 1575 1576 1577
}

//------------------------------inline_trig----------------------------------
// Inline sin/cos/tan instructions, if possible.  If rounding is required, do
// argument reduction which will turn into a fast/slow diamond.
bool LibraryCallKit::inline_trig(vmIntrinsics::ID id) {
1578 1579
  Node* arg = round_double_node(argument(0));
  Node* n = NULL;
D
duke 已提交
1580 1581

  switch (id) {
1582 1583 1584
  case vmIntrinsics::_dsin:  n = new (C) SinDNode(C, control(), arg);  break;
  case vmIntrinsics::_dcos:  n = new (C) CosDNode(C, control(), arg);  break;
  case vmIntrinsics::_dtan:  n = new (C) TanDNode(C, control(), arg);  break;
1585
  default:  fatal_unexpected_iid(id);  break;
D
duke 已提交
1586
  }
1587
  n = _gvn.transform(n);
D
duke 已提交
1588 1589

  // Rounding required?  Check for argument reduction!
1590
  if (Matcher::strict_fp_requires_explicit_rounding) {
D
duke 已提交
1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624
    static const double     pi_4 =  0.7853981633974483;
    static const double neg_pi_4 = -0.7853981633974483;
    // pi/2 in 80-bit extended precision
    // static const unsigned char pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0x3f,0x00,0x00,0x00,0x00,0x00,0x00};
    // -pi/2 in 80-bit extended precision
    // static const unsigned char neg_pi_2_bits_x[] = {0x35,0xc2,0x68,0x21,0xa2,0xda,0x0f,0xc9,0xff,0xbf,0x00,0x00,0x00,0x00,0x00,0x00};
    // Cutoff value for using this argument reduction technique
    //static const double    pi_2_minus_epsilon =  1.564660403643354;
    //static const double neg_pi_2_plus_epsilon = -1.564660403643354;

    // Pseudocode for sin:
    // if (x <= Math.PI / 4.0) {
    //   if (x >= -Math.PI / 4.0) return  fsin(x);
    //   if (x >= -Math.PI / 2.0) return -fcos(x + Math.PI / 2.0);
    // } else {
    //   if (x <=  Math.PI / 2.0) return  fcos(x - Math.PI / 2.0);
    // }
    // return StrictMath.sin(x);

    // Pseudocode for cos:
    // if (x <= Math.PI / 4.0) {
    //   if (x >= -Math.PI / 4.0) return  fcos(x);
    //   if (x >= -Math.PI / 2.0) return  fsin(x + Math.PI / 2.0);
    // } else {
    //   if (x <=  Math.PI / 2.0) return -fsin(x - Math.PI / 2.0);
    // }
    // return StrictMath.cos(x);

    // Actually, sticking in an 80-bit Intel value into C2 will be tough; it
    // requires a special machine instruction to load it.  Instead we'll try
    // the 'easy' case.  If we really need the extra range +/- PI/2 we'll
    // probably do the math inside the SIN encoding.

    // Make the merge point
1625 1626
    RegionNode* r = new (C) RegionNode(3);
    Node* phi = new (C) PhiNode(r, Type::DOUBLE);
D
duke 已提交
1627 1628

    // Flatten arg so we need only 1 test
1629
    Node *abs = _gvn.transform(new (C) AbsDNode(arg));
D
duke 已提交
1630 1631 1632
    // Node for PI/4 constant
    Node *pi4 = makecon(TypeD::make(pi_4));
    // Check PI/4 : abs(arg)
1633
    Node *cmp = _gvn.transform(new (C) CmpDNode(pi4,abs));
D
duke 已提交
1634
    // Check: If PI/4 < abs(arg) then go slow
1635
    Node *bol = _gvn.transform(new (C) BoolNode( cmp, BoolTest::lt ));
D
duke 已提交
1636 1637 1638 1639 1640
    // Branch either way
    IfNode *iff = create_and_xform_if(control(),bol, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
    set_control(opt_iff(r,iff));

    // Set fast path result
1641
    phi->init_req(2, n);
D
duke 已提交
1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662

    // Slow path - non-blocking leaf call
    Node* call = NULL;
    switch (id) {
    case vmIntrinsics::_dsin:
      call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
                               CAST_FROM_FN_PTR(address, SharedRuntime::dsin),
                               "Sin", NULL, arg, top());
      break;
    case vmIntrinsics::_dcos:
      call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
                               CAST_FROM_FN_PTR(address, SharedRuntime::dcos),
                               "Cos", NULL, arg, top());
      break;
    case vmIntrinsics::_dtan:
      call = make_runtime_call(RC_LEAF, OptoRuntime::Math_D_D_Type(),
                               CAST_FROM_FN_PTR(address, SharedRuntime::dtan),
                               "Tan", NULL, arg, top());
      break;
    }
    assert(control()->in(0) == call, "");
1663 1664 1665
    Node* slow_result = _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
    r->init_req(1, control());
    phi->init_req(1, slow_result);
D
duke 已提交
1666 1667 1668 1669

    // Post-merge
    set_control(_gvn.transform(r));
    record_for_igvn(r);
1670
    n = _gvn.transform(phi);
D
duke 已提交
1671 1672 1673

    C->set_has_split_ifs(true); // Has chance for split-if optimization
  }
1674
  set_result(n);
D
duke 已提交
1675 1676 1677
  return true;
}

1678 1679 1680 1681 1682
void LibraryCallKit::finish_pow_exp(Node* result, Node* x, Node* y, const TypeFunc* call_type, address funcAddr, const char* funcName) {
  //-------------------
  //result=(result.isNaN())? funcAddr():result;
  // Check: If isNaN() by checking result!=result? then either trap
  // or go to runtime
1683
  Node* cmpisnan = _gvn.transform(new (C) CmpDNode(result, result));
1684
  // Build the boolean node
1685
  Node* bolisnum = _gvn.transform(new (C) BoolNode(cmpisnan, BoolTest::eq));
1686 1687

  if (!too_many_traps(Deoptimization::Reason_intrinsic)) {
1688
    { BuildCutout unless(this, bolisnum, PROB_STATIC_FREQUENT);
1689 1690 1691 1692 1693
      // The pow or exp intrinsic returned a NaN, which requires a call
      // to the runtime.  Recompile with the runtime call.
      uncommon_trap(Deoptimization::Reason_intrinsic,
                    Deoptimization::Action_make_not_entrant);
    }
1694
    set_result(result);
1695 1696 1697 1698 1699
  } else {
    // If this inlining ever returned NaN in the past, we compile a call
    // to the runtime to properly handle corner cases

    IfNode* iff = create_and_xform_if(control(), bolisnum, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
1700 1701
    Node* if_slow = _gvn.transform(new (C) IfFalseNode(iff));
    Node* if_fast = _gvn.transform(new (C) IfTrueNode(iff));
1702 1703

    if (!if_slow->is_top()) {
1704
      RegionNode* result_region = new (C) RegionNode(3);
1705
      PhiNode*    result_val = new (C) PhiNode(result_region, Type::DOUBLE);
1706 1707 1708 1709 1710 1711 1712 1713 1714 1715

      result_region->init_req(1, if_fast);
      result_val->init_req(1, result);

      set_control(if_slow);

      const TypePtr* no_memory_effects = NULL;
      Node* rt = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
                                   no_memory_effects,
                                   x, top(), y, y ? top() : NULL);
1716
      Node* value = _gvn.transform(new (C) ProjNode(rt, TypeFunc::Parms+0));
1717
#ifdef ASSERT
1718
      Node* value_top = _gvn.transform(new (C) ProjNode(rt, TypeFunc::Parms+1));
1719 1720 1721 1722 1723
      assert(value_top == top(), "second value must be top");
#endif

      result_region->init_req(2, control());
      result_val->init_req(2, value);
1724
      set_result(result_region, result_val);
1725
    } else {
1726
      set_result(result);
1727 1728 1729 1730
    }
  }
}

D
duke 已提交
1731 1732 1733
//------------------------------inline_exp-------------------------------------
// Inline exp instructions, if possible.  The Intel hardware only misses
// really odd corner cases (+/- Infinity).  Just uncommon-trap them.
1734 1735
bool LibraryCallKit::inline_exp() {
  Node* arg = round_double_node(argument(0));
1736
  Node* n   = _gvn.transform(new (C) ExpDNode(C, control(), arg));
D
duke 已提交
1737

1738
  finish_pow_exp(n, arg, NULL, OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
D
duke 已提交
1739 1740 1741 1742 1743 1744 1745

  C->set_has_split_ifs(true); // Has chance for split-if optimization
  return true;
}

//------------------------------inline_pow-------------------------------------
// Inline power instructions, if possible.
1746
bool LibraryCallKit::inline_pow() {
D
duke 已提交
1747 1748
  // Pseudocode for pow
  // if (x <= 0.0) {
1749 1750 1751 1752
  //   long longy = (long)y;
  //   if ((double)longy == y) { // if y is long
  //     if (y + 1 == y) longy = 0; // huge number: even
  //     result = ((1&longy) == 0)?-DPow(abs(x), y):DPow(abs(x), y);
D
duke 已提交
1753 1754 1755 1756 1757 1758 1759
  //   } else {
  //     result = NaN;
  //   }
  // } else {
  //   result = DPow(x,y);
  // }
  // if (result != result)?  {
1760
  //   result = uncommon_trap() or runtime_call();
D
duke 已提交
1761 1762 1763
  // }
  // return result;

1764 1765
  Node* x = round_double_node(argument(0));
  Node* y = round_double_node(argument(2));
D
duke 已提交
1766

1767
  Node* result = NULL;
D
duke 已提交
1768

1769 1770
  if (!too_many_traps(Deoptimization::Reason_intrinsic)) {
    // Short form: skip the fancy tests and just check for NaN result.
1771
    result = _gvn.transform(new (C) PowDNode(C, control(), x, y));
D
duke 已提交
1772
  } else {
1773 1774
    // If this inlining ever returned NaN in the past, include all
    // checks + call to the runtime.
D
duke 已提交
1775 1776 1777

    // Set the merge point for If node with condition of (x <= 0.0)
    // There are four possible paths to region node and phi node
1778 1779
    RegionNode *r = new (C) RegionNode(4);
    Node *phi = new (C) PhiNode(r, Type::DOUBLE);
D
duke 已提交
1780 1781 1782 1783 1784

    // Build the first if node: if (x <= 0.0)
    // Node for 0 constant
    Node *zeronode = makecon(TypeD::ZERO);
    // Check x:0
1785
    Node *cmp = _gvn.transform(new (C) CmpDNode(x, zeronode));
D
duke 已提交
1786
    // Check: If (x<=0) then go complex path
1787
    Node *bol1 = _gvn.transform(new (C) BoolNode( cmp, BoolTest::le ));
D
duke 已提交
1788 1789 1790
    // Branch either way
    IfNode *if1 = create_and_xform_if(control(),bol1, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
    // Fast path taken; set region slot 3
1791
    Node *fast_taken = _gvn.transform(new (C) IfFalseNode(if1));
D
duke 已提交
1792 1793 1794
    r->init_req(3,fast_taken); // Capture fast-control

    // Fast path not-taken, i.e. slow path
1795
    Node *complex_path = _gvn.transform(new (C) IfTrueNode(if1));
D
duke 已提交
1796 1797

    // Set fast path result
1798
    Node *fast_result = _gvn.transform(new (C) PowDNode(C, control(), x, y));
D
duke 已提交
1799 1800 1801
    phi->init_req(3, fast_result);

    // Complex path
1802 1803
    // Build the second if node (if y is long)
    // Node for (long)y
1804
    Node *longy = _gvn.transform(new (C) ConvD2LNode(y));
1805
    // Node for (double)((long) y)
1806
    Node *doublelongy= _gvn.transform(new (C) ConvL2DNode(longy));
1807
    // Check (double)((long) y) : y
1808
    Node *cmplongy= _gvn.transform(new (C) CmpDNode(doublelongy, y));
1809 1810
    // Check if (y isn't long) then go to slow path

1811
    Node *bol2 = _gvn.transform(new (C) BoolNode( cmplongy, BoolTest::ne ));
T
twisti 已提交
1812
    // Branch either way
D
duke 已提交
1813
    IfNode *if2 = create_and_xform_if(complex_path,bol2, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
1814
    Node* ylong_path = _gvn.transform(new (C) IfFalseNode(if2));
1815

1816
    Node *slow_path = _gvn.transform(new (C) IfTrueNode(if2));
D
duke 已提交
1817

1818
    // Calculate DPow(abs(x), y)*(1 & (long)y)
D
duke 已提交
1819
    // Node for constant 1
1820 1821
    Node *conone = longcon(1);
    // 1& (long)y
1822
    Node *signnode= _gvn.transform(new (C) AndLNode(conone, longy));
1823 1824 1825 1826 1827 1828 1829

    // A huge number is always even. Detect a huge number by checking
    // if y + 1 == y and set integer to be tested for parity to 0.
    // Required for corner case:
    // (long)9.223372036854776E18 = max_jlong
    // (double)(long)9.223372036854776E18 = 9.223372036854776E18
    // max_jlong is odd but 9.223372036854776E18 is even
1830
    Node* yplus1 = _gvn.transform(new (C) AddDNode(y, makecon(TypeD::make(1))));
1831
    Node *cmpyplus1= _gvn.transform(new (C) CmpDNode(yplus1, y));
1832
    Node *bolyplus1 = _gvn.transform(new (C) BoolNode( cmpyplus1, BoolTest::eq ));
1833 1834 1835 1836 1837
    Node* correctedsign = NULL;
    if (ConditionalMoveLimit != 0) {
      correctedsign = _gvn.transform( CMoveNode::make(C, NULL, bolyplus1, signnode, longcon(0), TypeLong::LONG));
    } else {
      IfNode *ifyplus1 = create_and_xform_if(ylong_path,bolyplus1, PROB_FAIR, COUNT_UNKNOWN);
1838 1839
      RegionNode *r = new (C) RegionNode(3);
      Node *phi = new (C) PhiNode(r, TypeLong::LONG);
1840 1841
      r->init_req(1, _gvn.transform(new (C) IfFalseNode(ifyplus1)));
      r->init_req(2, _gvn.transform(new (C) IfTrueNode(ifyplus1)));
1842 1843 1844 1845 1846 1847 1848
      phi->init_req(1, signnode);
      phi->init_req(2, longcon(0));
      correctedsign = _gvn.transform(phi);
      ylong_path = _gvn.transform(r);
      record_for_igvn(r);
    }

D
duke 已提交
1849
    // zero node
1850 1851
    Node *conzero = longcon(0);
    // Check (1&(long)y)==0?
1852
    Node *cmpeq1 = _gvn.transform(new (C) CmpLNode(correctedsign, conzero));
1853
    // Check if (1&(long)y)!=0?, if so the result is negative
1854
    Node *bol3 = _gvn.transform(new (C) BoolNode( cmpeq1, BoolTest::ne ));
D
duke 已提交
1855
    // abs(x)
1856
    Node *absx=_gvn.transform(new (C) AbsDNode(x));
D
duke 已提交
1857
    // abs(x)^y
1858
    Node *absxpowy = _gvn.transform(new (C) PowDNode(C, control(), absx, y));
D
duke 已提交
1859
    // -abs(x)^y
1860
    Node *negabsxpowy = _gvn.transform(new (C) NegDNode (absxpowy));
1861 1862 1863 1864 1865 1866
    // (1&(long)y)==1?-DPow(abs(x), y):DPow(abs(x), y)
    Node *signresult = NULL;
    if (ConditionalMoveLimit != 0) {
      signresult = _gvn.transform( CMoveNode::make(C, NULL, bol3, absxpowy, negabsxpowy, Type::DOUBLE));
    } else {
      IfNode *ifyeven = create_and_xform_if(ylong_path,bol3, PROB_FAIR, COUNT_UNKNOWN);
1867 1868
      RegionNode *r = new (C) RegionNode(3);
      Node *phi = new (C) PhiNode(r, Type::DOUBLE);
1869 1870
      r->init_req(1, _gvn.transform(new (C) IfFalseNode(ifyeven)));
      r->init_req(2, _gvn.transform(new (C) IfTrueNode(ifyeven)));
1871 1872 1873 1874 1875 1876
      phi->init_req(1, absxpowy);
      phi->init_req(2, negabsxpowy);
      signresult = _gvn.transform(phi);
      ylong_path = _gvn.transform(r);
      record_for_igvn(r);
    }
D
duke 已提交
1877
    // Set complex path fast result
1878
    r->init_req(2, ylong_path);
D
duke 已提交
1879 1880 1881 1882 1883 1884 1885 1886 1887 1888
    phi->init_req(2, signresult);

    static const jlong nan_bits = CONST64(0x7ff8000000000000);
    Node *slow_result = makecon(TypeD::make(*(double*)&nan_bits)); // return NaN
    r->init_req(1,slow_path);
    phi->init_req(1,slow_result);

    // Post merge
    set_control(_gvn.transform(r));
    record_for_igvn(r);
1889
    result = _gvn.transform(phi);
D
duke 已提交
1890 1891
  }

1892
  finish_pow_exp(result, x, y, OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
D
duke 已提交
1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903

  C->set_has_split_ifs(true); // Has chance for split-if optimization
  return true;
}

//------------------------------runtime_math-----------------------------
bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
  assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
         "must be (DD)D or (D)D type");

  // Inputs
1904 1905
  Node* a = round_double_node(argument(0));
  Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : NULL;
D
duke 已提交
1906 1907 1908 1909 1910

  const TypePtr* no_memory_effects = NULL;
  Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
                                 no_memory_effects,
                                 a, top(), b, b ? top() : NULL);
1911
  Node* value = _gvn.transform(new (C) ProjNode(trig, TypeFunc::Parms+0));
D
duke 已提交
1912
#ifdef ASSERT
1913
  Node* value_top = _gvn.transform(new (C) ProjNode(trig, TypeFunc::Parms+1));
D
duke 已提交
1914 1915 1916
  assert(value_top == top(), "second value must be top");
#endif

1917
  set_result(value);
D
duke 已提交
1918 1919 1920 1921 1922
  return true;
}

//------------------------------inline_math_native-----------------------------
bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1923
#define FN_PTR(f) CAST_FROM_FN_PTR(address, f)
D
duke 已提交
1924 1925
  switch (id) {
    // These intrinsics are not properly supported on all hardware
1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936
  case vmIntrinsics::_dcos:   return Matcher::has_match_rule(Op_CosD)   ? inline_trig(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dcos),   "COS");
  case vmIntrinsics::_dsin:   return Matcher::has_match_rule(Op_SinD)   ? inline_trig(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dsin),   "SIN");
  case vmIntrinsics::_dtan:   return Matcher::has_match_rule(Op_TanD)   ? inline_trig(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dtan),   "TAN");

  case vmIntrinsics::_dlog:   return Matcher::has_match_rule(Op_LogD)   ? inline_math(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog),   "LOG");
  case vmIntrinsics::_dlog10: return Matcher::has_match_rule(Op_Log10D) ? inline_math(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), FN_PTR(SharedRuntime::dlog10), "LOG10");
D
duke 已提交
1937 1938

    // These intrinsics are supported on all hardware
1939 1940
  case vmIntrinsics::_dsqrt:  return Matcher::has_match_rule(Op_SqrtD)  ? inline_math(id) : false;
  case vmIntrinsics::_dabs:   return Matcher::has_match_rule(Op_AbsD)   ? inline_math(id) : false;
D
duke 已提交
1941

1942 1943 1944 1945 1946
  case vmIntrinsics::_dexp:   return Matcher::has_match_rule(Op_ExpD)   ? inline_exp()    :
    runtime_math(OptoRuntime::Math_D_D_Type(),  FN_PTR(SharedRuntime::dexp),  "EXP");
  case vmIntrinsics::_dpow:   return Matcher::has_match_rule(Op_PowD)   ? inline_pow()    :
    runtime_math(OptoRuntime::Math_DD_D_Type(), FN_PTR(SharedRuntime::dpow),  "POW");
#undef FN_PTR
D
duke 已提交
1947 1948 1949 1950 1951 1952

   // These intrinsics are not yet correctly implemented
  case vmIntrinsics::_datan2:
    return false;

  default:
1953
    fatal_unexpected_iid(id);
D
duke 已提交
1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967
    return false;
  }
}

static bool is_simple_name(Node* n) {
  return (n->req() == 1         // constant
          || (n->is_Type() && n->as_Type()->type()->singleton())
          || n->is_Proj()       // parameter or return value
          || n->is_Phi()        // local of some sort
          );
}

//----------------------------inline_min_max-----------------------------------
bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1968
  set_result(generate_min_max(id, argument(0), argument(1)));
D
duke 已提交
1969 1970 1971
  return true;
}

1972 1973 1974 1975 1976 1977 1978 1979
void LibraryCallKit::inline_math_mathExact(Node* math) {
  // If we didn't get the expected opcode it means we have optimized
  // the node to something else and don't need the exception edge.
  if (!math->is_MathExact()) {
    set_result(math);
    return;
  }

1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003
  Node* result = _gvn.transform( new(C) ProjNode(math, MathExactNode::result_proj_node));
  Node* flags = _gvn.transform( new(C) FlagsProjNode(math, MathExactNode::flags_proj_node));

  Node* bol = _gvn.transform( new (C) BoolNode(flags, BoolTest::overflow) );
  IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
  Node* fast_path = _gvn.transform( new (C) IfFalseNode(check));
  Node* slow_path = _gvn.transform( new (C) IfTrueNode(check) );

  {
    PreserveJVMState pjvms(this);
    PreserveReexecuteState preexecs(this);
    jvms()->set_should_reexecute(true);

    set_control(slow_path);
    set_i_o(i_o());

    uncommon_trap(Deoptimization::Reason_intrinsic,
                  Deoptimization::Action_none);
  }

  set_control(fast_path);
  set_result(result);
}

2004
bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2005
  Node* arg1 = argument(0);
2006 2007 2008
  Node* arg2 = NULL;

  if (is_increment) {
2009
    arg2 = intcon(1);
2010
  } else {
2011
    arg2 = argument(1);
2012
  }
2013 2014

  Node* add = _gvn.transform( new(C) AddExactINode(NULL, arg1, arg2) );
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041
  inline_math_mathExact(add);
  return true;
}

bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
  Node* arg1 = argument(0); // type long
  // argument(1) == TOP
  Node* arg2 = NULL;

  if (is_increment) {
    arg2 = longcon(1);
  } else {
    arg2 = argument(2); // type long
    // argument(3) == TOP
  }

  Node* add = _gvn.transform(new(C) AddExactLNode(NULL, arg1, arg2));
  inline_math_mathExact(add);
  return true;
}

bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
  Node* arg1 = argument(0);
  Node* arg2 = NULL;

  if (is_decrement) {
    arg2 = intcon(1);
2042
  } else {
2043
    arg2 = argument(1);
2044
  }
2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058

  Node* sub = _gvn.transform(new(C) SubExactINode(NULL, arg1, arg2));
  inline_math_mathExact(sub);
  return true;
}

bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
  Node* arg1 = argument(0); // type long
  // argument(1) == TOP
  Node* arg2 = NULL;

  if (is_decrement) {
    arg2 = longcon(1);
  } else {
2059
    arg2 = argument(2); // type long
2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101
    // argument(3) == TOP
  }

  Node* sub = _gvn.transform(new(C) SubExactLNode(NULL, arg1, arg2));
  inline_math_mathExact(sub);
  return true;
}

bool LibraryCallKit::inline_math_negateExactI() {
  Node* arg1 = argument(0);

  Node* neg = _gvn.transform(new(C) NegExactINode(NULL, arg1));
  inline_math_mathExact(neg);
  return true;
}

bool LibraryCallKit::inline_math_negateExactL() {
  Node* arg1 = argument(0);
  // argument(1) == TOP

  Node* neg = _gvn.transform(new(C) NegExactLNode(NULL, arg1));
  inline_math_mathExact(neg);
  return true;
}

bool LibraryCallKit::inline_math_multiplyExactI() {
  Node* arg1 = argument(0);
  Node* arg2 = argument(1);

  Node* mul = _gvn.transform(new(C) MulExactINode(NULL, arg1, arg2));
  inline_math_mathExact(mul);
  return true;
}

bool LibraryCallKit::inline_math_multiplyExactL() {
  Node* arg1 = argument(0);
  // argument(1) == TOP
  Node* arg2 = argument(2);
  // argument(3) == TOP

  Node* mul = _gvn.transform(new(C) MulExactLNode(NULL, arg1, arg2));
  inline_math_mathExact(mul);
2102 2103 2104
  return true;
}

D
duke 已提交
2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135
Node*
LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
  // These are the candidate return value:
  Node* xvalue = x0;
  Node* yvalue = y0;

  if (xvalue == yvalue) {
    return xvalue;
  }

  bool want_max = (id == vmIntrinsics::_max);

  const TypeInt* txvalue = _gvn.type(xvalue)->isa_int();
  const TypeInt* tyvalue = _gvn.type(yvalue)->isa_int();
  if (txvalue == NULL || tyvalue == NULL)  return top();
  // This is not really necessary, but it is consistent with a
  // hypothetical MaxINode::Value method:
  int widen = MAX2(txvalue->_widen, tyvalue->_widen);

  // %%% This folding logic should (ideally) be in a different place.
  // Some should be inside IfNode, and there to be a more reliable
  // transformation of ?: style patterns into cmoves.  We also want
  // more powerful optimizations around cmove and min/max.

  // Try to find a dominating comparison of these guys.
  // It can simplify the index computation for Arrays.copyOf
  // and similar uses of System.arraycopy.
  // First, compute the normalized version of CmpI(x, y).
  int   cmp_op = Op_CmpI;
  Node* xkey = xvalue;
  Node* ykey = yvalue;
2136
  Node* ideal_cmpxy = _gvn.transform(new(C) CmpINode(xkey, ykey));
D
duke 已提交
2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228
  if (ideal_cmpxy->is_Cmp()) {
    // E.g., if we have CmpI(length - offset, count),
    // it might idealize to CmpI(length, count + offset)
    cmp_op = ideal_cmpxy->Opcode();
    xkey = ideal_cmpxy->in(1);
    ykey = ideal_cmpxy->in(2);
  }

  // Start by locating any relevant comparisons.
  Node* start_from = (xkey->outcnt() < ykey->outcnt()) ? xkey : ykey;
  Node* cmpxy = NULL;
  Node* cmpyx = NULL;
  for (DUIterator_Fast kmax, k = start_from->fast_outs(kmax); k < kmax; k++) {
    Node* cmp = start_from->fast_out(k);
    if (cmp->outcnt() > 0 &&            // must have prior uses
        cmp->in(0) == NULL &&           // must be context-independent
        cmp->Opcode() == cmp_op) {      // right kind of compare
      if (cmp->in(1) == xkey && cmp->in(2) == ykey)  cmpxy = cmp;
      if (cmp->in(1) == ykey && cmp->in(2) == xkey)  cmpyx = cmp;
    }
  }

  const int NCMPS = 2;
  Node* cmps[NCMPS] = { cmpxy, cmpyx };
  int cmpn;
  for (cmpn = 0; cmpn < NCMPS; cmpn++) {
    if (cmps[cmpn] != NULL)  break;     // find a result
  }
  if (cmpn < NCMPS) {
    // Look for a dominating test that tells us the min and max.
    int depth = 0;                // Limit search depth for speed
    Node* dom = control();
    for (; dom != NULL; dom = IfNode::up_one_dom(dom, true)) {
      if (++depth >= 100)  break;
      Node* ifproj = dom;
      if (!ifproj->is_Proj())  continue;
      Node* iff = ifproj->in(0);
      if (!iff->is_If())  continue;
      Node* bol = iff->in(1);
      if (!bol->is_Bool())  continue;
      Node* cmp = bol->in(1);
      if (cmp == NULL)  continue;
      for (cmpn = 0; cmpn < NCMPS; cmpn++)
        if (cmps[cmpn] == cmp)  break;
      if (cmpn == NCMPS)  continue;
      BoolTest::mask btest = bol->as_Bool()->_test._test;
      if (ifproj->is_IfFalse())  btest = BoolTest(btest).negate();
      if (cmp->in(1) == ykey)    btest = BoolTest(btest).commute();
      // At this point, we know that 'x btest y' is true.
      switch (btest) {
      case BoolTest::eq:
        // They are proven equal, so we can collapse the min/max.
        // Either value is the answer.  Choose the simpler.
        if (is_simple_name(yvalue) && !is_simple_name(xvalue))
          return yvalue;
        return xvalue;
      case BoolTest::lt:          // x < y
      case BoolTest::le:          // x <= y
        return (want_max ? yvalue : xvalue);
      case BoolTest::gt:          // x > y
      case BoolTest::ge:          // x >= y
        return (want_max ? xvalue : yvalue);
      }
    }
  }

  // We failed to find a dominating test.
  // Let's pick a test that might GVN with prior tests.
  Node*          best_bol   = NULL;
  BoolTest::mask best_btest = BoolTest::illegal;
  for (cmpn = 0; cmpn < NCMPS; cmpn++) {
    Node* cmp = cmps[cmpn];
    if (cmp == NULL)  continue;
    for (DUIterator_Fast jmax, j = cmp->fast_outs(jmax); j < jmax; j++) {
      Node* bol = cmp->fast_out(j);
      if (!bol->is_Bool())  continue;
      BoolTest::mask btest = bol->as_Bool()->_test._test;
      if (btest == BoolTest::eq || btest == BoolTest::ne)  continue;
      if (cmp->in(1) == ykey)   btest = BoolTest(btest).commute();
      if (bol->outcnt() > (best_bol == NULL ? 0 : best_bol->outcnt())) {
        best_bol   = bol->as_Bool();
        best_btest = btest;
      }
    }
  }

  Node* answer_if_true  = NULL;
  Node* answer_if_false = NULL;
  switch (best_btest) {
  default:
    if (cmpxy == NULL)
      cmpxy = ideal_cmpxy;
2229
    best_bol = _gvn.transform(new(C) BoolNode(cmpxy, BoolTest::lt));
D
duke 已提交
2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288
    // and fall through:
  case BoolTest::lt:          // x < y
  case BoolTest::le:          // x <= y
    answer_if_true  = (want_max ? yvalue : xvalue);
    answer_if_false = (want_max ? xvalue : yvalue);
    break;
  case BoolTest::gt:          // x > y
  case BoolTest::ge:          // x >= y
    answer_if_true  = (want_max ? xvalue : yvalue);
    answer_if_false = (want_max ? yvalue : xvalue);
    break;
  }

  jint hi, lo;
  if (want_max) {
    // We can sharpen the minimum.
    hi = MAX2(txvalue->_hi, tyvalue->_hi);
    lo = MAX2(txvalue->_lo, tyvalue->_lo);
  } else {
    // We can sharpen the maximum.
    hi = MIN2(txvalue->_hi, tyvalue->_hi);
    lo = MIN2(txvalue->_lo, tyvalue->_lo);
  }

  // Use a flow-free graph structure, to avoid creating excess control edges
  // which could hinder other optimizations.
  // Since Math.min/max is often used with arraycopy, we want
  // tightly_coupled_allocation to be able to see beyond min/max expressions.
  Node* cmov = CMoveNode::make(C, NULL, best_bol,
                               answer_if_false, answer_if_true,
                               TypeInt::make(lo, hi, widen));

  return _gvn.transform(cmov);

  /*
  // This is not as desirable as it may seem, since Min and Max
  // nodes do not have a full set of optimizations.
  // And they would interfere, anyway, with 'if' optimizations
  // and with CMoveI canonical forms.
  switch (id) {
  case vmIntrinsics::_min:
    result_val = _gvn.transform(new (C, 3) MinINode(x,y)); break;
  case vmIntrinsics::_max:
    result_val = _gvn.transform(new (C, 3) MaxINode(x,y)); break;
  default:
    ShouldNotReachHere();
  }
  */
}

inline int
LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset) {
  const TypePtr* base_type = TypePtr::NULL_PTR;
  if (base != NULL)  base_type = _gvn.type(base)->isa_ptr();
  if (base_type == NULL) {
    // Unknown type.
    return Type::AnyPtr;
  } else if (base_type == TypePtr::NULL_PTR) {
    // Since this is a NULL+long form, we have to switch to a rawptr.
2289
    base   = _gvn.transform(new (C) CastX2PNode(offset));
D
duke 已提交
2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323
    offset = MakeConX(0);
    return Type::RawPtr;
  } else if (base_type->base() == Type::RawPtr) {
    return Type::RawPtr;
  } else if (base_type->isa_oopptr()) {
    // Base is never null => always a heap address.
    if (base_type->ptr() == TypePtr::NotNull) {
      return Type::OopPtr;
    }
    // Offset is small => always a heap address.
    const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
    if (offset_type != NULL &&
        base_type->offset() == 0 &&     // (should always be?)
        offset_type->_lo >= 0 &&
        !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
      return Type::OopPtr;
    }
    // Otherwise, it might either be oop+off or NULL+addr.
    return Type::AnyPtr;
  } else {
    // No information:
    return Type::AnyPtr;
  }
}

inline Node* LibraryCallKit::make_unsafe_address(Node* base, Node* offset) {
  int kind = classify_unsafe_addr(base, offset);
  if (kind == Type::RawPtr) {
    return basic_plus_adr(top(), base, offset);
  } else {
    return basic_plus_adr(base, offset);
  }
}

2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340
//--------------------------inline_number_methods-----------------------------
// inline int     Integer.numberOfLeadingZeros(int)
// inline int        Long.numberOfLeadingZeros(long)
//
// inline int     Integer.numberOfTrailingZeros(int)
// inline int        Long.numberOfTrailingZeros(long)
//
// inline int     Integer.bitCount(int)
// inline int        Long.bitCount(long)
//
// inline char  Character.reverseBytes(char)
// inline short     Short.reverseBytes(short)
// inline int     Integer.reverseBytes(int)
// inline long       Long.reverseBytes(long)
bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
  Node* arg = argument(0);
  Node* n;
D
duke 已提交
2341
  switch (id) {
2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354
  case vmIntrinsics::_numberOfLeadingZeros_i:   n = new (C) CountLeadingZerosINode( arg);  break;
  case vmIntrinsics::_numberOfLeadingZeros_l:   n = new (C) CountLeadingZerosLNode( arg);  break;
  case vmIntrinsics::_numberOfTrailingZeros_i:  n = new (C) CountTrailingZerosINode(arg);  break;
  case vmIntrinsics::_numberOfTrailingZeros_l:  n = new (C) CountTrailingZerosLNode(arg);  break;
  case vmIntrinsics::_bitCount_i:               n = new (C) PopCountINode(          arg);  break;
  case vmIntrinsics::_bitCount_l:               n = new (C) PopCountLNode(          arg);  break;
  case vmIntrinsics::_reverseBytes_c:           n = new (C) ReverseBytesUSNode(0,   arg);  break;
  case vmIntrinsics::_reverseBytes_s:           n = new (C) ReverseBytesSNode( 0,   arg);  break;
  case vmIntrinsics::_reverseBytes_i:           n = new (C) ReverseBytesINode( 0,   arg);  break;
  case vmIntrinsics::_reverseBytes_l:           n = new (C) ReverseBytesLNode( 0,   arg);  break;
  default:  fatal_unexpected_iid(id);  break;
  }
  set_result(_gvn.transform(n));
D
duke 已提交
2355 2356 2357 2358 2359 2360 2361
  return true;
}

//----------------------------inline_unsafe_access----------------------------

const static BasicType T_ADDRESS_HOLDER = T_LONG;

2362 2363
// Helper that guards and inserts a pre-barrier.
void LibraryCallKit::insert_pre_barrier(Node* base_oop, Node* offset,
2364
                                        Node* pre_val, bool need_mem_bar) {
2365 2366 2367 2368
  // We could be accessing the referent field of a reference object. If so, when G1
  // is enabled, we need to log the value in the referent field in an SATB buffer.
  // This routine performs some compile time filters and generates suitable
  // runtime filters that guard the pre-barrier code.
2369 2370 2371 2372
  // Also add memory barrier for non volatile load from the referent field
  // to prevent commoning of loads across safepoint.
  if (!UseG1GC && !need_mem_bar)
    return;
2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393

  // Some compile time checks.

  // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?
  const TypeX* otype = offset->find_intptr_t_type();
  if (otype != NULL && otype->is_con() &&
      otype->get_con() != java_lang_ref_Reference::referent_offset) {
    // Constant offset but not the reference_offset so just return
    return;
  }

  // We only need to generate the runtime guards for instances.
  const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();
  if (btype != NULL) {
    if (btype->isa_aryptr()) {
      // Array type so nothing to do
      return;
    }

    const TypeInstPtr* itype = btype->isa_instptr();
    if (itype != NULL) {
2394 2395
      // Can the klass of base_oop be statically determined to be
      // _not_ a sub-class of Reference and _not_ Object?
2396
      ciKlass* klass = itype->klass();
2397 2398 2399
      if ( klass->is_loaded() &&
          !klass->is_subtype_of(env()->Reference_klass()) &&
          !env()->Object_klass()->is_subtype_of(klass)) {
2400 2401 2402 2403 2404 2405 2406 2407 2408
        return;
      }
    }
  }

  // The compile time filters did not reject base_oop/offset so
  // we need to generate the following runtime filters
  //
  // if (offset == java_lang_ref_Reference::_reference_offset) {
2409 2410
  //   if (instance_of(base, java.lang.ref.Reference)) {
  //     pre_barrier(_, pre_val, ...);
2411 2412 2413
  //   }
  // }

2414 2415
  float likely   = PROB_LIKELY(  0.999);
  float unlikely = PROB_UNLIKELY(0.999);
2416

J
Merge  
johnc 已提交
2417
  IdealKit ideal(this);
2418 2419
#define __ ideal.

2420
  Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2421 2422 2423

  __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
      // Update graphKit memory and control from IdealKit.
J
Merge  
johnc 已提交
2424
      sync_kit(ideal);
2425 2426 2427 2428 2429

      Node* ref_klass_con = makecon(TypeKlassPtr::make(env()->Reference_klass()));
      Node* is_instof = gen_instanceof(base_oop, ref_klass_con);

      // Update IdealKit memory and control from graphKit.
J
Merge  
johnc 已提交
2430
      __ sync_kit(this);
2431 2432

      Node* one = __ ConI(1);
2433
      // is_instof == 0 if base_oop == NULL
2434 2435 2436
      __ if_then(is_instof, BoolTest::eq, one, unlikely); {

        // Update graphKit from IdeakKit.
J
Merge  
johnc 已提交
2437
        sync_kit(ideal);
2438 2439 2440 2441

        // Use the pre-barrier to record the value in the referent field
        pre_barrier(false /* do_load */,
                    __ ctrl(),
2442
                    NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2443 2444
                    pre_val /* pre_val */,
                    T_OBJECT);
2445 2446 2447 2448 2449
        if (need_mem_bar) {
          // Add memory barrier to prevent commoning reads from this field
          // across safepoint since GC can change its value.
          insert_mem_bar(Op_MemBarCPUOrder);
        }
2450
        // Update IdealKit from graphKit.
J
Merge  
johnc 已提交
2451
        __ sync_kit(this);
2452 2453 2454 2455 2456

      } __ end_if(); // _ref_type != ref_none
  } __ end_if(); // offset == referent_offset

  // Final sync IdealKit and GraphKit.
J
Merge  
johnc 已提交
2457
  final_sync(ideal);
2458 2459 2460 2461
#undef __
}


D
duke 已提交
2462 2463 2464
// Interpret Unsafe.fieldOffset cookies correctly:
extern jlong Unsafe_field_offset_to_byte_offset(jlong field_offset);

2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486
const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type, bool is_native_ptr) {
  // Attempt to infer a sharper value type from the offset and base type.
  ciKlass* sharpened_klass = NULL;

  // See if it is an instance field, with an object type.
  if (alias_type->field() != NULL) {
    assert(!is_native_ptr, "native pointer op cannot use a java address");
    if (alias_type->field()->type()->is_klass()) {
      sharpened_klass = alias_type->field()->type()->as_klass();
    }
  }

  // See if it is a narrow oop array.
  if (adr_type->isa_aryptr()) {
    if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
      const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
      if (elem_type != NULL) {
        sharpened_klass = elem_type->klass();
      }
    }
  }

2487 2488 2489
  // The sharpened class might be unloaded if there is no class loader
  // contraint in place.
  if (sharpened_klass != NULL && sharpened_klass->is_loaded()) {
2490 2491 2492
    const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);

#ifndef PRODUCT
2493
    if (C->print_intrinsics() || C->print_inlining()) {
2494 2495
      tty->print("  from base type: ");  adr_type->dump();
      tty->print("  sharpened value: ");  tjp->dump();
2496 2497 2498 2499 2500 2501 2502 2503
    }
#endif
    // Sharpen the value type.
    return tjp;
  }
  return NULL;
}

D
duke 已提交
2504 2505 2506 2507 2508 2509 2510
bool LibraryCallKit::inline_unsafe_access(bool is_native_ptr, bool is_store, BasicType type, bool is_volatile) {
  if (callee()->is_static())  return false;  // caller must have the capability!

#ifndef PRODUCT
  {
    ResourceMark rm;
    // Check the signatures.
2511
    ciSignature* sig = callee()->signature();
D
duke 已提交
2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548
#ifdef ASSERT
    if (!is_store) {
      // Object getObject(Object base, int/long offset), etc.
      BasicType rtype = sig->return_type()->basic_type();
      if (rtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::getAddress_name())
          rtype = T_ADDRESS;  // it is really a C void*
      assert(rtype == type, "getter must return the expected value");
      if (!is_native_ptr) {
        assert(sig->count() == 2, "oop getter has 2 arguments");
        assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
        assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
      } else {
        assert(sig->count() == 1, "native getter has 1 argument");
        assert(sig->type_at(0)->basic_type() == T_LONG, "getter base is long");
      }
    } else {
      // void putObject(Object base, int/long offset, Object x), etc.
      assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
      if (!is_native_ptr) {
        assert(sig->count() == 3, "oop putter has 3 arguments");
        assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
        assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
      } else {
        assert(sig->count() == 2, "native putter has 2 arguments");
        assert(sig->type_at(0)->basic_type() == T_LONG, "putter base is long");
      }
      BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
      if (vtype == T_ADDRESS_HOLDER && callee()->name() == ciSymbol::putAddress_name())
        vtype = T_ADDRESS;  // it is really a C void*
      assert(vtype == type, "putter must accept the expected value");
    }
#endif // ASSERT
 }
#endif //PRODUCT

  C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".

2549
  Node* receiver = argument(0);  // type: oop
D
duke 已提交
2550 2551

  // Build address expression.  See the code in inline_unsafe_prefetch.
2552 2553
  Node* adr;
  Node* heap_base_oop = top();
2554
  Node* offset = top();
2555
  Node* val;
2556

D
duke 已提交
2557 2558
  if (!is_native_ptr) {
    // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2559 2560 2561
    Node* base = argument(1);  // type: oop
    // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
    offset = argument(2);  // type: long
D
duke 已提交
2562 2563 2564 2565 2566 2567 2568 2569 2570
    // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
    // to be plain byte offsets, which are also the same as those accepted
    // by oopDesc::field_base.
    assert(Unsafe_field_offset_to_byte_offset(11) == 11,
           "fieldOffset must be byte-scaled");
    // 32-bit machines ignore the high half!
    offset = ConvL2X(offset);
    adr = make_unsafe_address(base, offset);
    heap_base_oop = base;
2571
    val = is_store ? argument(4) : NULL;
D
duke 已提交
2572
  } else {
2573 2574
    Node* ptr = argument(1);  // type: long
    ptr = ConvL2X(ptr);  // adjust Java long to machine word
D
duke 已提交
2575
    adr = make_unsafe_address(NULL, ptr);
2576
    val = is_store ? argument(3) : NULL;
D
duke 已提交
2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595
  }

  const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();

  // First guess at the value type.
  const Type *value_type = Type::get_const_basic_type(type);

  // Try to categorize the address.  If it comes up as TypeJavaPtr::BOTTOM,
  // there was not enough information to nail it down.
  Compile::AliasType* alias_type = C->alias_type(adr_type);
  assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");

  // We will need memory barriers unless we can determine a unique
  // alias category for this reference.  (Note:  If for some reason
  // the barriers get omitted and the unsafe reference begins to "pollute"
  // the alias analysis of the rest of the graph, either Compile::can_alias
  // or Compile::must_alias will throw a diagnostic assert.)
  bool need_mem_bar = (alias_type->adr_type() == TypeOopPtr::BOTTOM);

2596 2597 2598 2599
  // If we are reading the value of the referent field of a Reference
  // object (either by using Unsafe directly or through reflection)
  // then, if G1 is enabled, we need to record the referent in an
  // SATB log buffer using the pre-barrier mechanism.
2600 2601 2602
  // Also we need to add memory barrier to prevent commoning reads
  // from this field across safepoint since GC can change its value.
  bool need_read_barrier = !is_native_ptr && !is_store &&
2603 2604
                           offset != top() && heap_base_oop != top();

D
duke 已提交
2605
  if (!is_store && type == T_OBJECT) {
2606 2607
    const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type, is_native_ptr);
    if (tjp != NULL) {
D
duke 已提交
2608 2609 2610 2611
      value_type = tjp;
    }
  }

2612
  receiver = null_check(receiver);
D
duke 已提交
2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625
  if (stopped()) {
    return true;
  }
  // Heap pointers get a null-check from the interpreter,
  // as a courtesy.  However, this is not guaranteed by Unsafe,
  // and it is not possible to fully distinguish unintended nulls
  // from intended ones in this API.

  if (is_volatile) {
    // We need to emit leading and trailing CPU membars (see below) in
    // addition to memory membars when is_volatile. This is a little
    // too strong, but avoids the need to insert per-alias-type
    // volatile membars (for stores; compare Parse::do_put_xxx), which
T
twisti 已提交
2626
    // we cannot do effectively here because we probably only have a
D
duke 已提交
2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643
    // rough approximation of type.
    need_mem_bar = true;
    // For Stores, place a memory ordering barrier now.
    if (is_store)
      insert_mem_bar(Op_MemBarRelease);
  }

  // Memory barrier to prevent normal and 'unsafe' accesses from
  // bypassing each other.  Happens after null checks, so the
  // exception paths do not take memory state from the memory barrier,
  // so there's no problems making a strong assert about mixing users
  // of safe & unsafe memory.  Otherwise fails in a CTW of rt.jar
  // around 5701, class sun/reflect/UnsafeBooleanFieldAccessorImpl.
  if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);

  if (!is_store) {
    Node* p = make_load(control(), adr, value_type, type, adr_type, is_volatile);
2644
    // load value
D
duke 已提交
2645 2646 2647 2648 2649 2650
    switch (type) {
    case T_BOOLEAN:
    case T_CHAR:
    case T_BYTE:
    case T_SHORT:
    case T_INT:
2651
    case T_LONG:
D
duke 已提交
2652
    case T_FLOAT:
2653
    case T_DOUBLE:
2654
      break;
D
duke 已提交
2655
    case T_OBJECT:
2656
      if (need_read_barrier) {
2657
        insert_pre_barrier(heap_base_oop, offset, p, !(is_volatile || need_mem_bar));
2658
      }
D
duke 已提交
2659 2660 2661
      break;
    case T_ADDRESS:
      // Cast to an int type.
2662
      p = _gvn.transform(new (C) CastP2XNode(NULL, p));
D
duke 已提交
2663 2664
      p = ConvX2L(p);
      break;
2665 2666
    default:
      fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
D
duke 已提交
2667 2668
      break;
    }
2669 2670 2671 2672 2673
    // The load node has the control of the preceding MemBarCPUOrder.  All
    // following nodes will have the control of the MemBarCPUOrder inserted at
    // the end of this method.  So, pushing the load onto the stack at a later
    // point is fine.
    set_result(p);
D
duke 已提交
2674 2675 2676 2677 2678 2679 2680 2681 2682
  } else {
    // place effect of store into memory
    switch (type) {
    case T_DOUBLE:
      val = dstore_rounding(val);
      break;
    case T_ADDRESS:
      // Repackage the long as a pointer.
      val = ConvL2X(val);
2683
      val = _gvn.transform(new (C) CastX2PNode(val));
D
duke 已提交
2684 2685 2686 2687 2688 2689 2690 2691 2692
      break;
    }

    if (type != T_OBJECT ) {
      (void) store_to_memory(control(), adr, val, type, adr_type, is_volatile);
    } else {
      // Possibly an oop being stored to Java heap or native memory
      if (!TypePtr::NULL_PTR->higher_equal(_gvn.type(heap_base_oop))) {
        // oop to Java heap.
N
never 已提交
2693
        (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type);
D
duke 已提交
2694 2695 2696 2697 2698
      } else {
        // We can't tell at compile time if we are storing in the Java heap or outside
        // of it. So we need to emit code to conditionally do the proper type of
        // store.

2699
        IdealKit ideal(this);
2700
#define __ ideal.
D
duke 已提交
2701
        // QQQ who knows what probability is here??
2702 2703
        __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
          // Sync IdealKit and graphKit.
2704
          sync_kit(ideal);
2705 2706
          Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type);
          // Update IdealKit memory.
2707
          __ sync_kit(this);
2708 2709 2710 2711
        } __ else_(); {
          __ store(__ ctrl(), adr, val, type, alias_type->index(), is_volatile);
        } __ end_if();
        // Final sync IdealKit and GraphKit.
2712
        final_sync(ideal);
2713
#undef __
D
duke 已提交
2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736
      }
    }
  }

  if (is_volatile) {
    if (!is_store)
      insert_mem_bar(Op_MemBarAcquire);
    else
      insert_mem_bar(Op_MemBarVolatile);
  }

  if (need_mem_bar) insert_mem_bar(Op_MemBarCPUOrder);

  return true;
}

//----------------------------inline_unsafe_prefetch----------------------------

bool LibraryCallKit::inline_unsafe_prefetch(bool is_native_ptr, bool is_store, bool is_static) {
#ifndef PRODUCT
  {
    ResourceMark rm;
    // Check the signatures.
2737
    ciSignature* sig = callee()->signature();
D
duke 已提交
2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754
#ifdef ASSERT
    // Object getObject(Object base, int/long offset), etc.
    BasicType rtype = sig->return_type()->basic_type();
    if (!is_native_ptr) {
      assert(sig->count() == 2, "oop prefetch has 2 arguments");
      assert(sig->type_at(0)->basic_type() == T_OBJECT, "prefetch base is object");
      assert(sig->type_at(1)->basic_type() == T_LONG, "prefetcha offset is correct");
    } else {
      assert(sig->count() == 1, "native prefetch has 1 argument");
      assert(sig->type_at(0)->basic_type() == T_LONG, "prefetch base is long");
    }
#endif // ASSERT
  }
#endif // !PRODUCT

  C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".

2755 2756 2757 2758 2759 2760 2761
  const int idx = is_static ? 0 : 1;
  if (!is_static) {
    null_check_receiver();
    if (stopped()) {
      return true;
    }
  }
D
duke 已提交
2762 2763 2764 2765 2766

  // Build address expression.  See the code in inline_unsafe_access.
  Node *adr;
  if (!is_native_ptr) {
    // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2767 2768 2769
    Node* base   = argument(idx + 0);  // type: oop
    // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
    Node* offset = argument(idx + 1);  // type: long
D
duke 已提交
2770 2771 2772 2773 2774 2775 2776 2777 2778
    // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
    // to be plain byte offsets, which are also the same as those accepted
    // by oopDesc::field_base.
    assert(Unsafe_field_offset_to_byte_offset(11) == 11,
           "fieldOffset must be byte-scaled");
    // 32-bit machines ignore the high half!
    offset = ConvL2X(offset);
    adr = make_unsafe_address(base, offset);
  } else {
2779 2780
    Node* ptr = argument(idx + 0);  // type: long
    ptr = ConvL2X(ptr);  // adjust Java long to machine word
D
duke 已提交
2781 2782 2783 2784 2785 2786
    adr = make_unsafe_address(NULL, ptr);
  }

  // Generate the read or write prefetch
  Node *prefetch;
  if (is_store) {
2787
    prefetch = new (C) PrefetchWriteNode(i_o(), adr);
D
duke 已提交
2788
  } else {
2789
    prefetch = new (C) PrefetchReadNode(i_o(), adr);
D
duke 已提交
2790 2791 2792 2793 2794 2795 2796
  }
  prefetch->init_req(0, control());
  set_i_o(_gvn.transform(prefetch));

  return true;
}

2797
//----------------------------inline_unsafe_load_store----------------------------
2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813
// This method serves a couple of different customers (depending on LoadStoreKind):
//
// LS_cmpxchg:
//   public final native boolean compareAndSwapObject(Object o, long offset, Object expected, Object x);
//   public final native boolean compareAndSwapInt(   Object o, long offset, int    expected, int    x);
//   public final native boolean compareAndSwapLong(  Object o, long offset, long   expected, long   x);
//
// LS_xadd:
//   public int  getAndAddInt( Object o, long offset, int  delta)
//   public long getAndAddLong(Object o, long offset, long delta)
//
// LS_xchg:
//   int    getAndSet(Object o, long offset, int    newValue)
//   long   getAndSet(Object o, long offset, long   newValue)
//   Object getAndSet(Object o, long offset, Object newValue)
//
2814
bool LibraryCallKit::inline_unsafe_load_store(BasicType type, LoadStoreKind kind) {
D
duke 已提交
2815 2816 2817 2818 2819
  // This basic scheme here is the same as inline_unsafe_access, but
  // differs in enough details that combining them would make the code
  // overly confusing.  (This is a true fact! I originally combined
  // them, but even I was confused by it!) As much code/comments as
  // possible are retained from inline_unsafe_access though to make
T
twisti 已提交
2820
  // the correspondences clearer. - dl
D
duke 已提交
2821 2822 2823 2824

  if (callee()->is_static())  return false;  // caller must have the capability!

#ifndef PRODUCT
2825
  BasicType rtype;
D
duke 已提交
2826 2827
  {
    ResourceMark rm;
2828 2829
    // Check the signatures.
    ciSignature* sig = callee()->signature();
2830 2831 2832
    rtype = sig->return_type()->basic_type();
    if (kind == LS_xadd || kind == LS_xchg) {
      // Check the signatures.
D
duke 已提交
2833
#ifdef ASSERT
2834 2835 2836 2837 2838
      assert(rtype == type, "get and set must return the expected type");
      assert(sig->count() == 3, "get and set has 3 arguments");
      assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
      assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
      assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
D
duke 已提交
2839
#endif // ASSERT
2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850
    } else if (kind == LS_cmpxchg) {
      // Check the signatures.
#ifdef ASSERT
      assert(rtype == T_BOOLEAN, "CAS must return boolean");
      assert(sig->count() == 4, "CAS has 4 arguments");
      assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
      assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
#endif // ASSERT
    } else {
      ShouldNotReachHere();
    }
D
duke 已提交
2851 2852 2853 2854 2855
  }
#endif //PRODUCT

  C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".

2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878
  // Get arguments:
  Node* receiver = NULL;
  Node* base     = NULL;
  Node* offset   = NULL;
  Node* oldval   = NULL;
  Node* newval   = NULL;
  if (kind == LS_cmpxchg) {
    const bool two_slot_type = type2size[type] == 2;
    receiver = argument(0);  // type: oop
    base     = argument(1);  // type: oop
    offset   = argument(2);  // type: long
    oldval   = argument(4);  // type: oop, int, or long
    newval   = argument(two_slot_type ? 6 : 5);  // type: oop, int, or long
  } else if (kind == LS_xadd || kind == LS_xchg){
    receiver = argument(0);  // type: oop
    base     = argument(1);  // type: oop
    offset   = argument(2);  // type: long
    oldval   = NULL;
    newval   = argument(4);  // type: oop, int, or long
  }

  // Null check receiver.
  receiver = null_check(receiver);
D
duke 已提交
2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892
  if (stopped()) {
    return true;
  }

  // Build field offset expression.
  // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
  // to be plain byte offsets, which are also the same as those accepted
  // by oopDesc::field_base.
  assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
  // 32-bit machines ignore the high half of long offsets
  offset = ConvL2X(offset);
  Node* adr = make_unsafe_address(base, offset);
  const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();

2893 2894
  // For CAS, unlike inline_unsafe_access, there seems no point in
  // trying to refine types. Just use the coarse types here.
D
duke 已提交
2895 2896 2897
  const Type *value_type = Type::get_const_basic_type(type);
  Compile::AliasType* alias_type = C->alias_type(adr_type);
  assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2898 2899 2900 2901 2902 2903 2904 2905

  if (kind == LS_xchg && type == T_OBJECT) {
    const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
    if (tjp != NULL) {
      value_type = tjp;
    }
  }

D
duke 已提交
2906 2907
  int alias_idx = C->get_alias_index(adr_type);

2908 2909 2910
  // Memory-model-wise, a LoadStore acts like a little synchronized
  // block, so needs barriers on each side.  These don't translate
  // into actual barriers on most machines, but we still need rest of
D
duke 已提交
2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922
  // compiler to respect ordering.

  insert_mem_bar(Op_MemBarRelease);
  insert_mem_bar(Op_MemBarCPUOrder);

  // 4984716: MemBars must be inserted before this
  //          memory node in order to avoid a false
  //          dependency which will confuse the scheduler.
  Node *mem = memory(alias_idx);

  // For now, we handle only those cases that actually exist: ints,
  // longs, and Object. Adding others should be straightforward.
2923
  Node* load_store;
D
duke 已提交
2924 2925
  switch(type) {
  case T_INT:
2926
    if (kind == LS_xadd) {
2927
      load_store = _gvn.transform(new (C) GetAndAddINode(control(), mem, adr, newval, adr_type));
2928
    } else if (kind == LS_xchg) {
2929
      load_store = _gvn.transform(new (C) GetAndSetINode(control(), mem, adr, newval, adr_type));
2930
    } else if (kind == LS_cmpxchg) {
2931
      load_store = _gvn.transform(new (C) CompareAndSwapINode(control(), mem, adr, newval, oldval));
2932 2933 2934
    } else {
      ShouldNotReachHere();
    }
D
duke 已提交
2935 2936
    break;
  case T_LONG:
2937
    if (kind == LS_xadd) {
2938
      load_store = _gvn.transform(new (C) GetAndAddLNode(control(), mem, adr, newval, adr_type));
2939
    } else if (kind == LS_xchg) {
2940
      load_store = _gvn.transform(new (C) GetAndSetLNode(control(), mem, adr, newval, adr_type));
2941
    } else if (kind == LS_cmpxchg) {
2942
      load_store = _gvn.transform(new (C) CompareAndSwapLNode(control(), mem, adr, newval, oldval));
2943 2944 2945
    } else {
      ShouldNotReachHere();
    }
D
duke 已提交
2946 2947
    break;
  case T_OBJECT:
2948 2949 2950 2951 2952 2953 2954
    // Transformation of a value which could be NULL pointer (CastPP #NULL)
    // could be delayed during Parse (for example, in adjust_map_after_if()).
    // Execute transformation here to avoid barrier generation in such case.
    if (_gvn.type(newval) == TypePtr::NULL_PTR)
      newval = _gvn.makecon(TypePtr::NULL_PTR);

    // Reference stores need a store barrier.
2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976
    if (kind == LS_xchg) {
      // If pre-barrier must execute before the oop store, old value will require do_load here.
      if (!can_move_pre_barrier()) {
        pre_barrier(true /* do_load*/,
                    control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
                    NULL /* pre_val*/,
                    T_OBJECT);
      } // Else move pre_barrier to use load_store value, see below.
    } else if (kind == LS_cmpxchg) {
      // Same as for newval above:
      if (_gvn.type(oldval) == TypePtr::NULL_PTR) {
        oldval = _gvn.makecon(TypePtr::NULL_PTR);
      }
      // The only known value which might get overwritten is oldval.
      pre_barrier(false /* do_load */,
                  control(), NULL, NULL, max_juint, NULL, NULL,
                  oldval /* pre_val */,
                  T_OBJECT);
    } else {
      ShouldNotReachHere();
    }

2977
#ifdef _LP64
2978
    if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2979
      Node *newval_enc = _gvn.transform(new (C) EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
2980
      if (kind == LS_xchg) {
2981
        load_store = _gvn.transform(new (C) GetAndSetNNode(control(), mem, adr,
2982 2983 2984
                                                              newval_enc, adr_type, value_type->make_narrowoop()));
      } else {
        assert(kind == LS_cmpxchg, "wrong LoadStore operation");
2985 2986
        Node *oldval_enc = _gvn.transform(new (C) EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
        load_store = _gvn.transform(new (C) CompareAndSwapNNode(control(), mem, adr,
2987 2988
                                                                   newval_enc, oldval_enc));
      }
2989 2990
    } else
#endif
2991
    {
2992
      if (kind == LS_xchg) {
2993
        load_store = _gvn.transform(new (C) GetAndSetPNode(control(), mem, adr, newval, adr_type, value_type->is_oopptr()));
2994 2995
      } else {
        assert(kind == LS_cmpxchg, "wrong LoadStore operation");
2996
        load_store = _gvn.transform(new (C) CompareAndSwapPNode(control(), mem, adr, newval, oldval));
2997
      }
2998
    }
2999
    post_barrier(control(), load_store, base, adr, alias_idx, newval, T_OBJECT, true);
D
duke 已提交
3000 3001
    break;
  default:
3002
    fatal(err_msg_res("unexpected type %d: %s", type, type2name(type)));
D
duke 已提交
3003 3004 3005
    break;
  }

3006 3007 3008
  // SCMemProjNodes represent the memory state of a LoadStore. Their
  // main role is to prevent LoadStore nodes from being optimized away
  // when their results aren't used.
3009
  Node* proj = _gvn.transform(new (C) SCMemProjNode(load_store));
D
duke 已提交
3010 3011
  set_memory(proj, alias_idx);

3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028
  if (type == T_OBJECT && kind == LS_xchg) {
#ifdef _LP64
    if (adr->bottom_type()->is_ptr_to_narrowoop()) {
      load_store = _gvn.transform(new (C) DecodeNNode(load_store, load_store->get_ptr_type()));
    }
#endif
    if (can_move_pre_barrier()) {
      // Don't need to load pre_val. The old value is returned by load_store.
      // The pre_barrier can execute after the xchg as long as no safepoint
      // gets inserted between them.
      pre_barrier(false /* do_load */,
                  control(), NULL, NULL, max_juint, NULL, NULL,
                  load_store /* pre_val */,
                  T_OBJECT);
    }
  }

D
duke 已提交
3029 3030 3031 3032
  // Add the trailing membar surrounding the access
  insert_mem_bar(Op_MemBarCPUOrder);
  insert_mem_bar(Op_MemBarAcquire);

3033
  assert(type2size[load_store->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3034
  set_result(load_store);
D
duke 已提交
3035 3036 3037
  return true;
}

3038 3039 3040 3041
//----------------------------inline_unsafe_ordered_store----------------------
// public native void sun.misc.Unsafe.putOrderedObject(Object o, long offset, Object x);
// public native void sun.misc.Unsafe.putOrderedInt(Object o, long offset, int x);
// public native void sun.misc.Unsafe.putOrderedLong(Object o, long offset, long x);
D
duke 已提交
3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052
bool LibraryCallKit::inline_unsafe_ordered_store(BasicType type) {
  // This is another variant of inline_unsafe_access, differing in
  // that it always issues store-store ("release") barrier and ensures
  // store-atomicity (which only matters for "long").

  if (callee()->is_static())  return false;  // caller must have the capability!

#ifndef PRODUCT
  {
    ResourceMark rm;
    // Check the signatures.
3053
    ciSignature* sig = callee()->signature();
D
duke 已提交
3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065
#ifdef ASSERT
    BasicType rtype = sig->return_type()->basic_type();
    assert(rtype == T_VOID, "must return void");
    assert(sig->count() == 3, "has 3 arguments");
    assert(sig->type_at(0)->basic_type() == T_OBJECT, "base is object");
    assert(sig->type_at(1)->basic_type() == T_LONG, "offset is long");
#endif // ASSERT
  }
#endif //PRODUCT

  C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".

3066 3067 3068 3069 3070 3071 3072 3073
  // Get arguments:
  Node* receiver = argument(0);  // type: oop
  Node* base     = argument(1);  // type: oop
  Node* offset   = argument(2);  // type: long
  Node* val      = argument(4);  // type: oop, int, or long

  // Null check receiver.
  receiver = null_check(receiver);
D
duke 已提交
3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089
  if (stopped()) {
    return true;
  }

  // Build field offset expression.
  assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
  // 32-bit machines ignore the high half of long offsets
  offset = ConvL2X(offset);
  Node* adr = make_unsafe_address(base, offset);
  const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
  const Type *value_type = Type::get_const_basic_type(type);
  Compile::AliasType* alias_type = C->alias_type(adr_type);

  insert_mem_bar(Op_MemBarRelease);
  insert_mem_bar(Op_MemBarCPUOrder);
  // Ensure that the store is atomic for longs:
3090
  const bool require_atomic_access = true;
D
duke 已提交
3091 3092
  Node* store;
  if (type == T_OBJECT) // reference stores need a store barrier.
N
never 已提交
3093
    store = store_oop_to_unknown(control(), base, adr, adr_type, val, type);
D
duke 已提交
3094 3095 3096 3097 3098 3099 3100
  else {
    store = store_to_memory(control(), adr, val, type, adr_type, require_atomic_access);
  }
  insert_mem_bar(Op_MemBarCPUOrder);
  return true;
}

3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120
bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
  // Regardless of form, don't allow previous ld/st to move down,
  // then issue acquire, release, or volatile mem_bar.
  insert_mem_bar(Op_MemBarCPUOrder);
  switch(id) {
    case vmIntrinsics::_loadFence:
      insert_mem_bar(Op_MemBarAcquire);
      return true;
    case vmIntrinsics::_storeFence:
      insert_mem_bar(Op_MemBarRelease);
      return true;
    case vmIntrinsics::_fullFence:
      insert_mem_bar(Op_MemBarVolatile);
      return true;
    default:
      fatal_unexpected_iid(id);
      return false;
  }
}

R
rbackman 已提交
3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133
bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
  if (!kls->is_Con()) {
    return true;
  }
  const TypeKlassPtr* klsptr = kls->bottom_type()->isa_klassptr();
  if (klsptr == NULL) {
    return true;
  }
  ciInstanceKlass* ik = klsptr->klass()->as_instance_klass();
  // don't need a guard for a klass that is already initialized
  return !ik->is_initialized();
}

3134
//----------------------------inline_unsafe_allocate---------------------------
R
rbackman 已提交
3135
// public native Object sun.misc.Unsafe.allocateInstance(Class<?> cls);
D
duke 已提交
3136 3137
bool LibraryCallKit::inline_unsafe_allocate() {
  if (callee()->is_static())  return false;  // caller must have the capability!
3138 3139 3140

  null_check_receiver();  // null-check, then ignore
  Node* cls = null_check(argument(1));
D
duke 已提交
3141 3142
  if (stopped())  return true;

3143 3144
  Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
  kls = null_check(kls);
D
duke 已提交
3145 3146
  if (stopped())  return true;  // argument was like int.class

R
rbackman 已提交
3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159
  Node* test = NULL;
  if (LibraryCallKit::klass_needs_init_guard(kls)) {
    // Note:  The argument might still be an illegal value like
    // Serializable.class or Object[].class.   The runtime will handle it.
    // But we must make an explicit check for initialization.
    Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
    // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
    // can generate code to load it as unsigned byte.
    Node* inst = make_load(NULL, insp, TypeInt::UBYTE, T_BOOLEAN);
    Node* bits = intcon(InstanceKlass::fully_initialized);
    test = _gvn.transform(new (C) SubINode(inst, bits));
    // The 'test' is non-zero if we need to take a slow path.
  }
D
duke 已提交
3160 3161

  Node* obj = new_instance(kls, test);
3162
  set_result(obj);
D
duke 已提交
3163 3164 3165
  return true;
}

3166 3167 3168 3169 3170 3171 3172
#ifdef TRACE_HAVE_INTRINSICS
/*
 * oop -> myklass
 * myklass->trace_id |= USED
 * return myklass->trace_id & ~0x3
 */
bool LibraryCallKit::inline_native_classID() {
3173 3174 3175 3176
  null_check_receiver();  // null-check, then ignore
  Node* cls = null_check(argument(1), T_OBJECT);
  Node* kls = load_klass_from_mirror(cls, false, NULL, 0);
  kls = null_check(kls, T_OBJECT);
3177 3178 3179 3180
  ByteSize offset = TRACE_ID_OFFSET;
  Node* insp = basic_plus_adr(kls, in_bytes(offset));
  Node* tvalue = make_load(NULL, insp, TypeLong::LONG, T_LONG);
  Node* bits = longcon(~0x03l); // ignore bit 0 & 1
3181
  Node* andl = _gvn.transform(new (C) AndLNode(tvalue, bits));
3182
  Node* clsused = longcon(0x01l); // set the class bit
3183
  Node* orl = _gvn.transform(new (C) OrLNode(tvalue, clsused));
3184 3185 3186

  const TypePtr *adr_type = _gvn.type(insp)->isa_ptr();
  store_to_memory(control(), insp, orl, T_LONG, adr_type);
3187
  set_result(andl);
3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206
  return true;
}

bool LibraryCallKit::inline_native_threadID() {
  Node* tls_ptr = NULL;
  Node* cur_thr = generate_current_thread(tls_ptr);
  Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
  Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS);
  p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::thread_id_offset()));

  Node* threadid = NULL;
  size_t thread_id_size = OSThread::thread_id_size();
  if (thread_id_size == (size_t) BytesPerLong) {
    threadid = ConvL2I(make_load(control(), p, TypeLong::LONG, T_LONG));
  } else if (thread_id_size == (size_t) BytesPerInt) {
    threadid = make_load(control(), p, TypeInt::INT, T_INT);
  } else {
    ShouldNotReachHere();
  }
3207
  set_result(threadid);
3208 3209 3210 3211
  return true;
}
#endif

D
duke 已提交
3212 3213 3214
//------------------------inline_native_time_funcs--------------
// inline code for System.currentTimeMillis() and System.nanoTime()
// these have the same type and signature
3215
bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3216
  const TypeFunc* tf = OptoRuntime::void_long_Type();
D
duke 已提交
3217 3218
  const TypePtr* no_memory_effects = NULL;
  Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3219
  Node* value = _gvn.transform(new (C) ProjNode(time, TypeFunc::Parms+0));
D
duke 已提交
3220
#ifdef ASSERT
3221
  Node* value_top = _gvn.transform(new (C) ProjNode(time, TypeFunc::Parms+1));
D
duke 已提交
3222 3223
  assert(value_top == top(), "second value must be top");
#endif
3224
  set_result(value);
D
duke 已提交
3225 3226 3227 3228 3229 3230
  return true;
}

//------------------------inline_native_currentThread------------------
bool LibraryCallKit::inline_native_currentThread() {
  Node* junk = NULL;
3231
  set_result(generate_current_thread(junk));
D
duke 已提交
3232 3233 3234 3235
  return true;
}

//------------------------inline_native_isInterrupted------------------
3236
// private native boolean java.lang.Thread.isInterrupted(boolean ClearInterrupted);
D
duke 已提交
3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248
bool LibraryCallKit::inline_native_isInterrupted() {
  // Add a fast path to t.isInterrupted(clear_int):
  //   (t == Thread.current() && (!TLS._osthread._interrupted || !clear_int))
  //   ? TLS._osthread._interrupted : /*slow path:*/ t.isInterrupted(clear_int)
  // So, in the common case that the interrupt bit is false,
  // we avoid making a call into the VM.  Even if the interrupt bit
  // is true, if the clear_int argument is false, we avoid the VM call.
  // However, if the receiver is not currentThread, we must call the VM,
  // because there must be some locking done around the operation.

  // We only go to the fast case code if we pass two guards.
  // Paths which do not pass are accumulated in the slow_region.
3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263

  enum {
    no_int_result_path   = 1, // t == Thread.current() && !TLS._osthread._interrupted
    no_clear_result_path = 2, // t == Thread.current() &&  TLS._osthread._interrupted && !clear_int
    slow_result_path     = 3, // slow path: t.isInterrupted(clear_int)
    PATH_LIMIT
  };

  // Ensure that it's not possible to move the load of TLS._osthread._interrupted flag
  // out of the function.
  insert_mem_bar(Op_MemBarCPUOrder);

  RegionNode* result_rgn = new (C) RegionNode(PATH_LIMIT);
  PhiNode*    result_val = new (C) PhiNode(result_rgn, TypeInt::BOOL);

3264
  RegionNode* slow_region = new (C) RegionNode(1);
D
duke 已提交
3265 3266 3267 3268 3269 3270
  record_for_igvn(slow_region);

  // (a) Receiving thread must be the current thread.
  Node* rec_thr = argument(0);
  Node* tls_ptr = NULL;
  Node* cur_thr = generate_current_thread(tls_ptr);
3271 3272
  Node* cmp_thr = _gvn.transform(new (C) CmpPNode(cur_thr, rec_thr));
  Node* bol_thr = _gvn.transform(new (C) BoolNode(cmp_thr, BoolTest::ne));
D
duke 已提交
3273

3274
  generate_slow_guard(bol_thr, slow_region);
D
duke 已提交
3275 3276 3277 3278 3279

  // (b) Interrupt bit on TLS must be false.
  Node* p = basic_plus_adr(top()/*!oop*/, tls_ptr, in_bytes(JavaThread::osthread_offset()));
  Node* osthread = make_load(NULL, p, TypeRawPtr::NOTNULL, T_ADDRESS);
  p = basic_plus_adr(top()/*!oop*/, osthread, in_bytes(OSThread::interrupted_offset()));
3280

3281 3282
  // Set the control input on the field _interrupted read to prevent it floating up.
  Node* int_bit = make_load(control(), p, TypeInt::BOOL, T_INT);
3283 3284
  Node* cmp_bit = _gvn.transform(new (C) CmpINode(int_bit, intcon(0)));
  Node* bol_bit = _gvn.transform(new (C) BoolNode(cmp_bit, BoolTest::ne));
D
duke 已提交
3285 3286 3287 3288

  IfNode* iff_bit = create_and_map_if(control(), bol_bit, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);

  // First fast path:  if (!TLS._interrupted) return false;
3289
  Node* false_bit = _gvn.transform(new (C) IfFalseNode(iff_bit));
D
duke 已提交
3290 3291 3292 3293
  result_rgn->init_req(no_int_result_path, false_bit);
  result_val->init_req(no_int_result_path, intcon(0));

  // drop through to next case
3294
  set_control( _gvn.transform(new (C) IfTrueNode(iff_bit)));
D
duke 已提交
3295 3296 3297

  // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
  Node* clr_arg = argument(1);
3298 3299
  Node* cmp_arg = _gvn.transform(new (C) CmpINode(clr_arg, intcon(0)));
  Node* bol_arg = _gvn.transform(new (C) BoolNode(cmp_arg, BoolTest::ne));
D
duke 已提交
3300 3301 3302
  IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);

  // Second fast path:  ... else if (!clear_int) return true;
3303
  Node* false_arg = _gvn.transform(new (C) IfFalseNode(iff_arg));
D
duke 已提交
3304 3305 3306 3307
  result_rgn->init_req(no_clear_result_path, false_arg);
  result_val->init_req(no_clear_result_path, intcon(1));

  // drop through to next case
3308
  set_control( _gvn.transform(new (C) IfTrueNode(iff_arg)));
D
duke 已提交
3309 3310 3311

  // (d) Otherwise, go to the slow path.
  slow_region->add_req(control());
3312
  set_control( _gvn.transform(slow_region));
D
duke 已提交
3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326

  if (stopped()) {
    // There is no slow path.
    result_rgn->init_req(slow_result_path, top());
    result_val->init_req(slow_result_path, top());
  } else {
    // non-virtual because it is a private non-static
    CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_isInterrupted);

    Node* slow_val = set_results_for_java_call(slow_call);
    // this->control() comes from set_results_for_java_call

    Node* fast_io  = slow_call->in(TypeFunc::I_O);
    Node* fast_mem = slow_call->in(TypeFunc::Memory);
3327

D
duke 已提交
3328
    // These two phis are pre-filled with copies of of the fast IO and Memory
3329 3330
    PhiNode* result_mem  = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);
    PhiNode* result_io   = PhiNode::make(result_rgn, fast_io,  Type::ABIO);
D
duke 已提交
3331 3332

    result_rgn->init_req(slow_result_path, control());
3333 3334
    result_io ->init_req(slow_result_path, i_o());
    result_mem->init_req(slow_result_path, reset_memory());
D
duke 已提交
3335 3336
    result_val->init_req(slow_result_path, slow_val);

3337 3338
    set_all_memory(_gvn.transform(result_mem));
    set_i_o(       _gvn.transform(result_io));
D
duke 已提交
3339 3340 3341
  }

  C->set_has_split_ifs(true); // Has chance for split-if optimization
3342
  set_result(result_rgn, result_val);
D
duke 已提交
3343 3344 3345 3346 3347 3348
  return true;
}

//---------------------------load_mirror_from_klass----------------------------
// Given a klass oop, load its java mirror (a java.lang.Class oop).
Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3349
  Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
D
duke 已提交
3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367
  return make_load(NULL, p, TypeInstPtr::MIRROR, T_OBJECT);
}

//-----------------------load_klass_from_mirror_common-------------------------
// Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
// Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
// and branch to the given path on the region.
// If never_see_null, take an uncommon trap on null, so we can optimistically
// compile for the non-null case.
// If the region is NULL, force never_see_null = true.
Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
                                                    bool never_see_null,
                                                    RegionNode* region,
                                                    int null_path,
                                                    int offset) {
  if (region == NULL)  never_see_null = true;
  Node* p = basic_plus_adr(mirror, offset);
  const TypeKlassPtr*  kls_type = TypeKlassPtr::OBJECT_OR_NULL;
3368
  Node* kls = _gvn.transform( LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
D
duke 已提交
3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385
  Node* null_ctl = top();
  kls = null_check_oop(kls, &null_ctl, never_see_null);
  if (region != NULL) {
    // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
    region->init_req(null_path, null_ctl);
  } else {
    assert(null_ctl == top(), "no loose ends");
  }
  return kls;
}

//--------------------(inline_native_Class_query helpers)---------------------
// Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE, JVM_ACC_HAS_FINALIZER.
// Fall through if (mods & mask) == bits, take the guard otherwise.
Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
  // Branch around if the given klass has the given modifier bit set.
  // Like generate_guard, adds a new path onto the region.
3386
  Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
D
duke 已提交
3387 3388 3389
  Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT);
  Node* mask = intcon(modifier_mask);
  Node* bits = intcon(modifier_bits);
3390 3391 3392
  Node* mbit = _gvn.transform(new (C) AndINode(mods, mask));
  Node* cmp  = _gvn.transform(new (C) CmpINode(mbit, bits));
  Node* bol  = _gvn.transform(new (C) BoolNode(cmp, BoolTest::ne));
D
duke 已提交
3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407
  return generate_fair_guard(bol, region);
}
Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
  return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
}

//-------------------------inline_native_Class_query-------------------
bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
  const Type* return_type = TypeInt::BOOL;
  Node* prim_return_value = top();  // what happens if it's a primitive class?
  bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
  bool expect_prim = false;     // most of these guys expect to work on refs

  enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };

3408 3409 3410
  Node* mirror = argument(0);
  Node* obj    = top();

D
duke 已提交
3411 3412 3413 3414
  switch (id) {
  case vmIntrinsics::_isInstance:
    // nothing is an instance of a primitive type
    prim_return_value = intcon(0);
3415
    obj = argument(1);
D
duke 已提交
3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445
    break;
  case vmIntrinsics::_getModifiers:
    prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
    assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
    return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
    break;
  case vmIntrinsics::_isInterface:
    prim_return_value = intcon(0);
    break;
  case vmIntrinsics::_isArray:
    prim_return_value = intcon(0);
    expect_prim = true;  // cf. ObjectStreamClass.getClassSignature
    break;
  case vmIntrinsics::_isPrimitive:
    prim_return_value = intcon(1);
    expect_prim = true;  // obviously
    break;
  case vmIntrinsics::_getSuperclass:
    prim_return_value = null();
    return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
    break;
  case vmIntrinsics::_getComponentType:
    prim_return_value = null();
    return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
    break;
  case vmIntrinsics::_getClassAccessFlags:
    prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
    return_type = TypeInt::INT;  // not bool!  6297094
    break;
  default:
3446 3447
    fatal_unexpected_iid(id);
    break;
D
duke 已提交
3448 3449 3450 3451 3452 3453
  }

  const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
  if (mirror_con == NULL)  return false;  // cannot happen?

#ifndef PRODUCT
3454
  if (C->print_intrinsics() || C->print_inlining()) {
D
duke 已提交
3455 3456 3457 3458 3459 3460 3461 3462 3463 3464
    ciType* k = mirror_con->java_mirror_type();
    if (k) {
      tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
      k->print_name();
      tty->cr();
    }
  }
#endif

  // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3465
  RegionNode* region = new (C) RegionNode(PATH_LIMIT);
D
duke 已提交
3466
  record_for_igvn(region);
3467
  PhiNode* phi = new (C) PhiNode(region, return_type);
D
duke 已提交
3468 3469 3470 3471 3472 3473 3474 3475

  // The mirror will never be null of Reflection.getClassAccessFlags, however
  // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
  // if it is. See bug 4774291.

  // For Reflection.getClassAccessFlags(), the null check occurs in
  // the wrong place; see inline_unsafe_access(), above, for a similar
  // situation.
3476
  mirror = null_check(mirror);
D
duke 已提交
3477 3478 3479 3480 3481 3482 3483
  // If mirror or obj is dead, only null-path is taken.
  if (stopped())  return true;

  if (expect_prim)  never_see_null = false;  // expect nulls (meaning prims)

  // Now load the mirror's klass metaobject, and null-check it.
  // Side-effects region with the control path if the klass is null.
3484
  Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
D
duke 已提交
3485 3486
  // If kls is null, we have a primitive mirror.
  phi->init_req(_prim_path, prim_return_value);
3487
  if (stopped()) { set_result(region, phi); return true; }
3488
  bool safe_for_replace = (region->in(_prim_path) == top());
D
duke 已提交
3489 3490 3491 3492 3493 3494 3495 3496 3497 3498

  Node* p;  // handy temp
  Node* null_ctl;

  // Now that we have the non-null klass, we can perform the real query.
  // For constant classes, the query will constant-fold in LoadNode::Value.
  Node* query_value = top();
  switch (id) {
  case vmIntrinsics::_isInstance:
    // nothing is an instance of a primitive type
3499
    query_value = gen_instanceof(obj, kls, safe_for_replace);
D
duke 已提交
3500 3501 3502
    break;

  case vmIntrinsics::_getModifiers:
3503
    p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
D
duke 已提交
3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542
    query_value = make_load(NULL, p, TypeInt::INT, T_INT);
    break;

  case vmIntrinsics::_isInterface:
    // (To verify this code sequence, check the asserts in JVM_IsInterface.)
    if (generate_interface_guard(kls, region) != NULL)
      // A guard was added.  If the guard is taken, it was an interface.
      phi->add_req(intcon(1));
    // If we fall through, it's a plain class.
    query_value = intcon(0);
    break;

  case vmIntrinsics::_isArray:
    // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
    if (generate_array_guard(kls, region) != NULL)
      // A guard was added.  If the guard is taken, it was an array.
      phi->add_req(intcon(1));
    // If we fall through, it's a plain class.
    query_value = intcon(0);
    break;

  case vmIntrinsics::_isPrimitive:
    query_value = intcon(0); // "normal" path produces false
    break;

  case vmIntrinsics::_getSuperclass:
    // The rules here are somewhat unfortunate, but we can still do better
    // with random logic than with a JNI call.
    // Interfaces store null or Object as _super, but must report null.
    // Arrays store an intermediate super as _super, but must report Object.
    // Other types can report the actual _super.
    // (To verify this code sequence, check the asserts in JVM_IsInterface.)
    if (generate_interface_guard(kls, region) != NULL)
      // A guard was added.  If the guard is taken, it was an interface.
      phi->add_req(null());
    if (generate_array_guard(kls, region) != NULL)
      // A guard was added.  If the guard is taken, it was an array.
      phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
    // If we fall through, it's a plain class.  Get its _super.
3543
    p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3544
    kls = _gvn.transform( LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL));
D
duke 已提交
3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560
    null_ctl = top();
    kls = null_check_oop(kls, &null_ctl);
    if (null_ctl != top()) {
      // If the guard is taken, Object.superClass is null (both klass and mirror).
      region->add_req(null_ctl);
      phi   ->add_req(null());
    }
    if (!stopped()) {
      query_value = load_mirror_from_klass(kls);
    }
    break;

  case vmIntrinsics::_getComponentType:
    if (generate_array_guard(kls, region) != NULL) {
      // Be sure to pin the oop load to the guard edge just created:
      Node* is_array_ctrl = region->in(region->req()-1);
3561
      Node* cma = basic_plus_adr(kls, in_bytes(ArrayKlass::component_mirror_offset()));
D
duke 已提交
3562 3563 3564 3565 3566 3567 3568
      Node* cmo = make_load(is_array_ctrl, cma, TypeInstPtr::MIRROR, T_OBJECT);
      phi->add_req(cmo);
    }
    query_value = null();  // non-array case is null
    break;

  case vmIntrinsics::_getClassAccessFlags:
3569
    p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
D
duke 已提交
3570 3571 3572 3573
    query_value = make_load(NULL, p, TypeInt::INT, T_INT);
    break;

  default:
3574 3575
    fatal_unexpected_iid(id);
    break;
D
duke 已提交
3576 3577 3578 3579 3580 3581 3582
  }

  // Fall-through is the normal case of a query to a real class.
  phi->init_req(1, query_value);
  region->init_req(1, control());

  C->set_has_split_ifs(true); // Has chance for split-if optimization
3583
  set_result(region, phi);
D
duke 已提交
3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608
  return true;
}

//--------------------------inline_native_subtype_check------------------------
// This intrinsic takes the JNI calls out of the heart of
// UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
bool LibraryCallKit::inline_native_subtype_check() {
  // Pull both arguments off the stack.
  Node* args[2];                // two java.lang.Class mirrors: superc, subc
  args[0] = argument(0);
  args[1] = argument(1);
  Node* klasses[2];             // corresponding Klasses: superk, subk
  klasses[0] = klasses[1] = top();

  enum {
    // A full decision tree on {superc is prim, subc is prim}:
    _prim_0_path = 1,           // {P,N} => false
                                // {P,P} & superc!=subc => false
    _prim_same_path,            // {P,P} & superc==subc => true
    _prim_1_path,               // {N,P} => false
    _ref_subtype_path,          // {N,N} & subtype check wins => true
    _both_ref_path,             // {N,N} & subtype check loses => false
    PATH_LIMIT
  };

3609 3610
  RegionNode* region = new (C) RegionNode(PATH_LIMIT);
  Node*       phi    = new (C) PhiNode(region, TypeInt::BOOL);
D
duke 已提交
3611 3612 3613 3614 3615 3616 3617 3618 3619 3620
  record_for_igvn(region);

  const TypePtr* adr_type = TypeRawPtr::BOTTOM;   // memory type of loads
  const TypeKlassPtr* kls_type = TypeKlassPtr::OBJECT_OR_NULL;
  int class_klass_offset = java_lang_Class::klass_offset_in_bytes();

  // First null-check both mirrors and load each mirror's klass metaobject.
  int which_arg;
  for (which_arg = 0; which_arg <= 1; which_arg++) {
    Node* arg = args[which_arg];
3621
    arg = null_check(arg);
D
duke 已提交
3622
    if (stopped())  break;
3623
    args[which_arg] = arg;
D
duke 已提交
3624 3625

    Node* p = basic_plus_adr(arg, class_klass_offset);
3626
    Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
D
duke 已提交
3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656
    klasses[which_arg] = _gvn.transform(kls);
  }

  // Having loaded both klasses, test each for null.
  bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
  for (which_arg = 0; which_arg <= 1; which_arg++) {
    Node* kls = klasses[which_arg];
    Node* null_ctl = top();
    kls = null_check_oop(kls, &null_ctl, never_see_null);
    int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
    region->init_req(prim_path, null_ctl);
    if (stopped())  break;
    klasses[which_arg] = kls;
  }

  if (!stopped()) {
    // now we have two reference types, in klasses[0..1]
    Node* subk   = klasses[1];  // the argument to isAssignableFrom
    Node* superk = klasses[0];  // the receiver
    region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
    // now we have a successful reference subtype check
    region->set_req(_ref_subtype_path, control());
  }

  // If both operands are primitive (both klasses null), then
  // we must return true when they are identical primitives.
  // It is convenient to test this after the first null klass check.
  set_control(region->in(_prim_0_path)); // go back to first null check
  if (!stopped()) {
    // Since superc is primitive, make a guard for the superc==subc case.
3657 3658
    Node* cmp_eq = _gvn.transform(new (C) CmpPNode(args[0], args[1]));
    Node* bol_eq = _gvn.transform(new (C) BoolNode(cmp_eq, BoolTest::eq));
D
duke 已提交
3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684
    generate_guard(bol_eq, region, PROB_FAIR);
    if (region->req() == PATH_LIMIT+1) {
      // A guard was added.  If the added guard is taken, superc==subc.
      region->swap_edges(PATH_LIMIT, _prim_same_path);
      region->del_req(PATH_LIMIT);
    }
    region->set_req(_prim_0_path, control()); // Not equal after all.
  }

  // these are the only paths that produce 'true':
  phi->set_req(_prim_same_path,   intcon(1));
  phi->set_req(_ref_subtype_path, intcon(1));

  // pull together the cases:
  assert(region->req() == PATH_LIMIT, "sane region");
  for (uint i = 1; i < region->req(); i++) {
    Node* ctl = region->in(i);
    if (ctl == NULL || ctl == top()) {
      region->set_req(i, top());
      phi   ->set_req(i, top());
    } else if (phi->in(i) == NULL) {
      phi->set_req(i, intcon(0)); // all other paths produce 'false'
    }
  }

  set_control(_gvn.transform(region));
3685
  set_result(_gvn.transform(phi));
D
duke 已提交
3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706
  return true;
}

//---------------------generate_array_guard_common------------------------
Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
                                                  bool obj_array, bool not_array) {
  // If obj_array/non_array==false/false:
  // Branch around if the given klass is in fact an array (either obj or prim).
  // If obj_array/non_array==false/true:
  // Branch around if the given klass is not an array klass of any kind.
  // If obj_array/non_array==true/true:
  // Branch around if the kls is not an oop array (kls is int[], String, etc.)
  // If obj_array/non_array==true/false:
  // Branch around if the kls is an oop array (Object[] or subtype)
  //
  // Like generate_guard, adds a new path onto the region.
  jint  layout_con = 0;
  Node* layout_val = get_layout_helper(kls, layout_con);
  if (layout_val == NULL) {
    bool query = (obj_array
                  ? Klass::layout_helper_is_objArray(layout_con)
3707
                  : Klass::layout_helper_is_array(layout_con));
D
duke 已提交
3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722
    if (query == not_array) {
      return NULL;                       // never a branch
    } else {                             // always a branch
      Node* always_branch = control();
      if (region != NULL)
        region->add_req(always_branch);
      set_control(top());
      return always_branch;
    }
  }
  // Now test the correct condition.
  jint  nval = (obj_array
                ? ((jint)Klass::_lh_array_tag_type_value
                   <<    Klass::_lh_array_tag_shift)
                : Klass::_lh_neutral_value);
3723
  Node* cmp = _gvn.transform(new(C) CmpINode(layout_val, intcon(nval)));
D
duke 已提交
3724 3725 3726
  BoolTest::mask btest = BoolTest::lt;  // correct for testing is_[obj]array
  // invert the test if we are looking for a non-array
  if (not_array)  btest = BoolTest(btest).negate();
3727
  Node* bol = _gvn.transform(new(C) BoolNode(cmp, btest));
D
duke 已提交
3728 3729 3730 3731 3732
  return generate_fair_guard(bol, region);
}


//-----------------------inline_native_newArray--------------------------
3733
// private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
D
duke 已提交
3734 3735 3736 3737
bool LibraryCallKit::inline_native_newArray() {
  Node* mirror    = argument(0);
  Node* count_val = argument(1);

3738
  mirror = null_check(mirror);
3739 3740
  // If mirror or obj is dead, only null-path is taken.
  if (stopped())  return true;
D
duke 已提交
3741 3742

  enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
3743 3744 3745 3746 3747 3748
  RegionNode* result_reg = new(C) RegionNode(PATH_LIMIT);
  PhiNode*    result_val = new(C) PhiNode(result_reg,
                                          TypeInstPtr::NOTNULL);
  PhiNode*    result_io  = new(C) PhiNode(result_reg, Type::ABIO);
  PhiNode*    result_mem = new(C) PhiNode(result_reg, Type::MEMORY,
                                          TypePtr::BOTTOM);
D
duke 已提交
3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777

  bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
  Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
                                                  result_reg, _slow_path);
  Node* normal_ctl   = control();
  Node* no_array_ctl = result_reg->in(_slow_path);

  // Generate code for the slow case.  We make a call to newArray().
  set_control(no_array_ctl);
  if (!stopped()) {
    // Either the input type is void.class, or else the
    // array klass has not yet been cached.  Either the
    // ensuing call will throw an exception, or else it
    // will cache the array klass for next time.
    PreserveJVMState pjvms(this);
    CallJavaNode* slow_call = generate_method_call_static(vmIntrinsics::_newArray);
    Node* slow_result = set_results_for_java_call(slow_call);
    // this->control() comes from set_results_for_java_call
    result_reg->set_req(_slow_path, control());
    result_val->set_req(_slow_path, slow_result);
    result_io ->set_req(_slow_path, i_o());
    result_mem->set_req(_slow_path, reset_memory());
  }

  set_control(normal_ctl);
  if (!stopped()) {
    // Normal case:  The array type has been cached in the java.lang.Class.
    // The following call works fine even if the array type is polymorphic.
    // It could be a dynamic mix of int[], boolean[], Object[], etc.
3778
    Node* obj = new_array(klass_node, count_val, 0);  // no arguments to push
D
duke 已提交
3779 3780 3781 3782 3783 3784 3785 3786
    result_reg->init_req(_normal_path, control());
    result_val->init_req(_normal_path, obj);
    result_io ->init_req(_normal_path, i_o());
    result_mem->init_req(_normal_path, reset_memory());
  }

  // Return the combined state.
  set_i_o(        _gvn.transform(result_io)  );
3787
  set_all_memory( _gvn.transform(result_mem));
D
duke 已提交
3788

3789 3790
  C->set_has_split_ifs(true); // Has chance for split-if optimization
  set_result(result_reg, result_val);
D
duke 已提交
3791 3792 3793 3794
  return true;
}

//----------------------inline_native_getLength--------------------------
3795
// public static native int java.lang.reflect.Array.getLength(Object array);
D
duke 已提交
3796 3797 3798
bool LibraryCallKit::inline_native_getLength() {
  if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;

3799
  Node* array = null_check(argument(0));
D
duke 已提交
3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817
  // If array is dead, only null-path is taken.
  if (stopped())  return true;

  // Deoptimize if it is a non-array.
  Node* non_array = generate_non_array_guard(load_object_klass(array), NULL);

  if (non_array != NULL) {
    PreserveJVMState pjvms(this);
    set_control(non_array);
    uncommon_trap(Deoptimization::Reason_intrinsic,
                  Deoptimization::Action_maybe_recompile);
  }

  // If control is dead, only non-array-path is taken.
  if (stopped())  return true;

  // The works fine even if the array type is polymorphic.
  // It could be a dynamic mix of int[], boolean[], Object[], etc.
3818
  Node* result = load_array_length(array);
D
duke 已提交
3819

3820 3821
  C->set_has_split_ifs(true);  // Has chance for split-if optimization
  set_result(result);
D
duke 已提交
3822 3823 3824 3825
  return true;
}

//------------------------inline_array_copyOf----------------------------
3826 3827
// public static <T,U> T[] java.util.Arrays.copyOf(     U[] original, int newLength,         Class<? extends T[]> newType);
// public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from,      int to, Class<? extends T[]> newType);
D
duke 已提交
3828 3829 3830
bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
  if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;

3831
  // Get the arguments.
D
duke 已提交
3832 3833 3834 3835 3836
  Node* original          = argument(0);
  Node* start             = is_copyOfRange? argument(1): intcon(0);
  Node* end               = is_copyOfRange? argument(2): argument(1);
  Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);

3837
  Node* newcopy;
D
duke 已提交
3838

3839 3840
  // Set the original stack and the reexecute bit for the interpreter to reexecute
  // the bytecode that invokes Arrays.copyOf if deoptimization happens.
3841 3842
  { PreserveReexecuteState preexecs(this);
    jvms()->set_should_reexecute(true);
D
duke 已提交
3843

3844 3845
    array_type_mirror = null_check(array_type_mirror);
    original          = null_check(original);
D
duke 已提交
3846

3847 3848
    // Check if a null path was taken unconditionally.
    if (stopped())  return true;
D
duke 已提交
3849

3850
    Node* orig_length = load_array_length(original);
D
duke 已提交
3851

3852 3853
    Node* klass_node = load_klass_from_mirror(array_type_mirror, false, NULL, 0);
    klass_node = null_check(klass_node);
D
duke 已提交
3854

3855
    RegionNode* bailout = new (C) RegionNode(1);
3856
    record_for_igvn(bailout);
D
duke 已提交
3857

3858 3859 3860 3861 3862 3863 3864
    // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
    // Bail out if that is so.
    Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
    if (not_objArray != NULL) {
      // Improve the klass node's type from the new optimistic assumption:
      ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
      const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
3865
      Node* cast = new (C) CastPPNode(klass_node, akls);
3866 3867 3868
      cast->init_req(0, control());
      klass_node = _gvn.transform(cast);
    }
D
duke 已提交
3869

3870 3871 3872
    // Bail out if either start or end is negative.
    generate_negative_guard(start, bailout, &start);
    generate_negative_guard(end,   bailout, &end);
D
duke 已提交
3873

3874 3875
    Node* length = end;
    if (_gvn.type(start) != TypeInt::ZERO) {
3876
      length = _gvn.transform(new (C) SubINode(end, start));
3877 3878 3879
    }

    // Bail out if length is negative.
3880 3881 3882 3883
    // Without this the new_array would throw
    // NegativeArraySizeException but IllegalArgumentException is what
    // should be thrown
    generate_negative_guard(length, bailout, &length);
3884 3885 3886

    if (bailout->req() > 1) {
      PreserveJVMState pjvms(this);
3887
      set_control(_gvn.transform(bailout));
3888 3889 3890 3891 3892
      uncommon_trap(Deoptimization::Reason_intrinsic,
                    Deoptimization::Action_maybe_recompile);
    }

    if (!stopped()) {
3893 3894
      // How many elements will we copy from the original?
      // The answer is MinI(orig_length - start, length).
3895
      Node* orig_tail = _gvn.transform(new (C) SubINode(orig_length, start));
3896 3897
      Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);

3898
      newcopy = new_array(klass_node, length, 0);  // no argments to push
3899 3900 3901 3902 3903 3904 3905

      // Generate a direct call to the right arraycopy function(s).
      // We know the copy is disjoint but we might not know if the
      // oop stores need checking.
      // Extreme case:  Arrays.copyOf((Integer[])x, 10, String[].class).
      // This will fail a store-check if x contains any non-nulls.
      bool disjoint_bases = true;
3906 3907 3908
      // if start > orig_length then the length of the copy may be
      // negative.
      bool length_never_negative = !is_copyOfRange;
3909 3910 3911
      generate_arraycopy(TypeAryPtr::OOPS, T_OBJECT,
                         original, start, newcopy, intcon(0), moved,
                         disjoint_bases, length_never_negative);
3912
    }
3913
  } // original reexecute is set back here
D
duke 已提交
3914 3915

  C->set_has_split_ifs(true); // Has chance for split-if optimization
3916 3917 3918
  if (!stopped()) {
    set_result(newcopy);
  }
D
duke 已提交
3919 3920 3921 3922 3923 3924 3925 3926 3927 3928
  return true;
}


//----------------------generate_virtual_guard---------------------------
// Helper for hashCode and clone.  Peeks inside the vtable to avoid a call.
Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
                                             RegionNode* slow_region) {
  ciMethod* method = callee();
  int vtable_index = method->vtable_index();
3929 3930
  assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
         err_msg_res("bad index %d", vtable_index));
3931 3932
  // Get the Method* out of the appropriate vtable entry.
  int entry_offset  = (InstanceKlass::vtable_start_offset() +
D
duke 已提交
3933 3934 3935
                     vtable_index*vtableEntry::size()) * wordSize +
                     vtableEntry::method_offset_in_bytes();
  Node* entry_addr  = basic_plus_adr(obj_klass, entry_offset);
3936
  Node* target_call = make_load(NULL, entry_addr, TypePtr::NOTNULL, T_ADDRESS);
D
duke 已提交
3937 3938

  // Compare the target method with the expected method (e.g., Object.hashCode).
3939
  const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
D
duke 已提交
3940 3941

  Node* native_call = makecon(native_call_addr);
3942 3943
  Node* chk_native  = _gvn.transform(new(C) CmpPNode(target_call, native_call));
  Node* test_native = _gvn.transform(new(C) BoolNode(chk_native, BoolTest::ne));
D
duke 已提交
3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967

  return generate_slow_guard(test_native, slow_region);
}

//-----------------------generate_method_call----------------------------
// Use generate_method_call to make a slow-call to the real
// method if the fast path fails.  An alternative would be to
// use a stub like OptoRuntime::slow_arraycopy_Java.
// This only works for expanding the current library call,
// not another intrinsic.  (E.g., don't use this for making an
// arraycopy call inside of the copyOf intrinsic.)
CallJavaNode*
LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
  // When compiling the intrinsic method itself, do not use this technique.
  guarantee(callee() != C->method(), "cannot make slow-call to self");

  ciMethod* method = callee();
  // ensure the JVMS we have will be correct for this call
  guarantee(method_id == method->intrinsic_id(), "must match");

  const TypeFunc* tf = TypeFunc::make(method);
  CallJavaNode* slow_call;
  if (is_static) {
    assert(!is_virtual, "");
3968
    slow_call = new(C) CallStaticJavaNode(C, tf,
3969 3970
                           SharedRuntime::get_resolve_static_call_stub(),
                           method, bci());
D
duke 已提交
3971
  } else if (is_virtual) {
3972
    null_check_receiver();
3973
    int vtable_index = Method::invalid_vtable_index;
D
duke 已提交
3974 3975 3976 3977 3978 3979 3980
    if (UseInlineCaches) {
      // Suppress the vtable call
    } else {
      // hashCode and clone are not a miranda methods,
      // so the vtable index is fixed.
      // No need to use the linkResolver to get it.
       vtable_index = method->vtable_index();
3981 3982
       assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
              err_msg_res("bad index %d", vtable_index));
D
duke 已提交
3983
    }
3984 3985 3986
    slow_call = new(C) CallDynamicJavaNode(tf,
                          SharedRuntime::get_resolve_virtual_call_stub(),
                          method, vtable_index, bci());
D
duke 已提交
3987
  } else {  // neither virtual nor static:  opt_virtual
3988
    null_check_receiver();
3989
    slow_call = new(C) CallStaticJavaNode(C, tf,
D
duke 已提交
3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007
                                SharedRuntime::get_resolve_opt_virtual_call_stub(),
                                method, bci());
    slow_call->set_optimized_virtual(true);
  }
  set_arguments_for_java_call(slow_call);
  set_edges_for_java_call(slow_call);
  return slow_call;
}


//------------------------------inline_native_hashcode--------------------
// Build special case code for calls to hashCode on an object.
bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
  assert(is_static == callee()->is_static(), "correct intrinsic selection");
  assert(!(is_virtual && is_static), "either virtual, special, or static");

  enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };

4008 4009 4010 4011 4012 4013
  RegionNode* result_reg = new(C) RegionNode(PATH_LIMIT);
  PhiNode*    result_val = new(C) PhiNode(result_reg,
                                          TypeInt::INT);
  PhiNode*    result_io  = new(C) PhiNode(result_reg, Type::ABIO);
  PhiNode*    result_mem = new(C) PhiNode(result_reg, Type::MEMORY,
                                          TypePtr::BOTTOM);
D
duke 已提交
4014 4015 4016
  Node* obj = NULL;
  if (!is_static) {
    // Check for hashing null object
4017
    obj = null_check_receiver();
D
duke 已提交
4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032
    if (stopped())  return true;        // unconditionally null
    result_reg->init_req(_null_path, top());
    result_val->init_req(_null_path, top());
  } else {
    // Do a null check, and return zero if null.
    // System.identityHashCode(null) == 0
    obj = argument(0);
    Node* null_ctl = top();
    obj = null_check_oop(obj, &null_ctl);
    result_reg->init_req(_null_path, null_ctl);
    result_val->init_req(_null_path, _gvn.intcon(0));
  }

  // Unconditionally null?  Then return right away.
  if (stopped()) {
4033
    set_control( result_reg->in(_null_path));
D
duke 已提交
4034
    if (!stopped())
4035
      set_result(result_val->in(_null_path));
D
duke 已提交
4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046
    return true;
  }

  // After null check, get the object's klass.
  Node* obj_klass = load_object_klass(obj);

  // This call may be virtual (invokevirtual) or bound (invokespecial).
  // For each case we generate slightly different code.

  // We only go to the fast case code if we pass a number of guards.  The
  // paths which do not pass are accumulated in the slow_region.
4047
  RegionNode* slow_region = new (C) RegionNode(1);
D
duke 已提交
4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061
  record_for_igvn(slow_region);

  // If this is a virtual call, we generate a funny guard.  We pull out
  // the vtable entry corresponding to hashCode() from the target object.
  // If the target method which we are calling happens to be the native
  // Object hashCode() method, we pass the guard.  We do not need this
  // guard for non-virtual calls -- the caller is known to be the native
  // Object hashCode().
  if (is_virtual) {
    generate_virtual_guard(obj_klass, slow_region);
  }

  // Get the header out of the object, use LoadMarkNode when available
  Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4062
  Node* header = make_load(control(), header_addr, TypeX_X, TypeX_X->basic_type());
D
duke 已提交
4063 4064 4065

  // Test the header to see if it is unlocked.
  Node *lock_mask      = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
4066
  Node *lmasked_header = _gvn.transform(new (C) AndXNode(header, lock_mask));
D
duke 已提交
4067
  Node *unlocked_val   = _gvn.MakeConX(markOopDesc::unlocked_value);
4068 4069
  Node *chk_unlocked   = _gvn.transform(new (C) CmpXNode( lmasked_header, unlocked_val));
  Node *test_unlocked  = _gvn.transform(new (C) BoolNode( chk_unlocked, BoolTest::ne));
D
duke 已提交
4070 4071 4072 4073 4074 4075 4076 4077 4078

  generate_slow_guard(test_unlocked, slow_region);

  // Get the hash value and check to see that it has been properly assigned.
  // We depend on hash_mask being at most 32 bits and avoid the use of
  // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
  // vm: see markOop.hpp.
  Node *hash_mask      = _gvn.intcon(markOopDesc::hash_mask);
  Node *hash_shift     = _gvn.intcon(markOopDesc::hash_shift);
4079
  Node *hshifted_header= _gvn.transform(new (C) URShiftXNode(header, hash_shift));
D
duke 已提交
4080
  // This hack lets the hash bits live anywhere in the mark object now, as long
T
twisti 已提交
4081
  // as the shift drops the relevant bits into the low 32 bits.  Note that
D
duke 已提交
4082 4083 4084
  // Java spec says that HashCode is an int so there's no point in capturing
  // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
  hshifted_header      = ConvX2I(hshifted_header);
4085
  Node *hash_val       = _gvn.transform(new (C) AndINode(hshifted_header, hash_mask));
D
duke 已提交
4086 4087

  Node *no_hash_val    = _gvn.intcon(markOopDesc::no_hash);
4088 4089
  Node *chk_assigned   = _gvn.transform(new (C) CmpINode( hash_val, no_hash_val));
  Node *test_assigned  = _gvn.transform(new (C) BoolNode( chk_assigned, BoolTest::eq));
D
duke 已提交
4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107

  generate_slow_guard(test_assigned, slow_region);

  Node* init_mem = reset_memory();
  // fill in the rest of the null path:
  result_io ->init_req(_null_path, i_o());
  result_mem->init_req(_null_path, init_mem);

  result_val->init_req(_fast_path, hash_val);
  result_reg->init_req(_fast_path, control());
  result_io ->init_req(_fast_path, i_o());
  result_mem->init_req(_fast_path, init_mem);

  // Generate code for the slow case.  We make a call to hashCode().
  set_control(_gvn.transform(slow_region));
  if (!stopped()) {
    // No need for PreserveJVMState, because we're using up the present state.
    set_all_memory(init_mem);
4108
    vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
D
duke 已提交
4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119
    CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
    Node* slow_result = set_results_for_java_call(slow_call);
    // this->control() comes from set_results_for_java_call
    result_reg->init_req(_slow_path, control());
    result_val->init_req(_slow_path, slow_result);
    result_io  ->set_req(_slow_path, i_o());
    result_mem ->set_req(_slow_path, reset_memory());
  }

  // Return the combined state.
  set_i_o(        _gvn.transform(result_io)  );
4120
  set_all_memory( _gvn.transform(result_mem));
D
duke 已提交
4121

4122
  set_result(result_reg, result_val);
D
duke 已提交
4123 4124 4125 4126
  return true;
}

//---------------------------inline_native_getClass----------------------------
4127 4128
// public final native Class<?> java.lang.Object.getClass();
//
T
twisti 已提交
4129
// Build special case code for calls to getClass on an object.
D
duke 已提交
4130
bool LibraryCallKit::inline_native_getClass() {
4131
  Node* obj = null_check_receiver();
D
duke 已提交
4132
  if (stopped())  return true;
4133
  set_result(load_mirror_from_klass(load_object_klass(obj)));
D
duke 已提交
4134 4135 4136 4137
  return true;
}

//-----------------inline_native_Reflection_getCallerClass---------------------
4138
// public static native Class<?> sun.reflect.Reflection.getCallerClass();
4139
//
D
duke 已提交
4140 4141
// In the presence of deep enough inlining, getCallerClass() becomes a no-op.
//
4142 4143 4144
// NOTE: This code must perform the same logic as JVM_GetCallerClass
// in that it must skip particular security frames and checks for
// caller sensitive methods.
D
duke 已提交
4145 4146
bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
#ifndef PRODUCT
4147
  if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
D
duke 已提交
4148 4149 4150 4151 4152 4153
    tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
  }
#endif

  if (!jvms()->has_method()) {
#ifndef PRODUCT
4154
    if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
D
duke 已提交
4155 4156 4157 4158 4159 4160 4161
      tty->print_cr("  Bailing out because intrinsic was inlined at top level");
    }
#endif
    return false;
  }

  // Walk back up the JVM state to find the caller at the required
4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177
  // depth.
  JVMState* caller_jvms = jvms();

  // Cf. JVM_GetCallerClass
  // NOTE: Start the loop at depth 1 because the current JVM state does
  // not include the Reflection.getCallerClass() frame.
  for (int n = 1; caller_jvms != NULL; caller_jvms = caller_jvms->caller(), n++) {
    ciMethod* m = caller_jvms->method();
    switch (n) {
    case 0:
      fatal("current JVM state does not include the Reflection.getCallerClass frame");
      break;
    case 1:
      // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
      if (!m->caller_sensitive()) {
#ifndef PRODUCT
4178
        if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4179 4180 4181 4182
          tty->print_cr("  Bailing out: CallerSensitive annotation expected at frame %d", n);
        }
#endif
        return false;  // bail-out; let JVM_GetCallerClass do the work
D
duke 已提交
4183
      }
4184 4185 4186 4187 4188 4189 4190 4191
      break;
    default:
      if (!m->is_ignored_by_security_stack_walk()) {
        // We have reached the desired frame; return the holder class.
        // Acquire method holder as java.lang.Class and push as constant.
        ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
        ciInstance* caller_mirror = caller_klass->java_mirror();
        set_result(makecon(TypeInstPtr::make(caller_mirror)));
D
duke 已提交
4192 4193

#ifndef PRODUCT
4194
        if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4195 4196 4197 4198 4199 4200 4201 4202 4203
          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());
          tty->print_cr("  JVM state at this point:");
          for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
            ciMethod* m = jvms()->of_depth(i)->method();
            tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
          }
        }
#endif
        return true;
D
duke 已提交
4204
      }
4205
      break;
D
duke 已提交
4206 4207 4208 4209
    }
  }

#ifndef PRODUCT
4210
  if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4211
    tty->print_cr("  Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
D
duke 已提交
4212
    tty->print_cr("  JVM state at this point:");
4213
    for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4214
      ciMethod* m = jvms()->of_depth(i)->method();
4215
      tty->print_cr("   %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
D
duke 已提交
4216 4217 4218 4219
    }
  }
#endif

4220
  return false;  // bail-out; let JVM_GetCallerClass do the work
D
duke 已提交
4221 4222 4223
}

bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4224 4225
  Node* arg = argument(0);
  Node* result;
D
duke 已提交
4226 4227

  switch (id) {
4228 4229 4230 4231
  case vmIntrinsics::_floatToRawIntBits:    result = new (C) MoveF2INode(arg);  break;
  case vmIntrinsics::_intBitsToFloat:       result = new (C) MoveI2FNode(arg);  break;
  case vmIntrinsics::_doubleToRawLongBits:  result = new (C) MoveD2LNode(arg);  break;
  case vmIntrinsics::_longBitsToDouble:     result = new (C) MoveL2DNode(arg);  break;
D
duke 已提交
4232 4233 4234

  case vmIntrinsics::_doubleToLongBits: {
    // two paths (plus control) merge in a wood
4235 4236
    RegionNode *r = new (C) RegionNode(3);
    Node *phi = new (C) PhiNode(r, TypeLong::LONG);
D
duke 已提交
4237

4238
    Node *cmpisnan = _gvn.transform(new (C) CmpDNode(arg, arg));
D
duke 已提交
4239
    // Build the boolean node
4240
    Node *bolisnan = _gvn.transform(new (C) BoolNode(cmpisnan, BoolTest::ne));
D
duke 已提交
4241 4242 4243 4244 4245 4246 4247

    // Branch either way.
    // NaN case is less traveled, which makes all the difference.
    IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
    Node *opt_isnan = _gvn.transform(ifisnan);
    assert( opt_isnan->is_If(), "Expect an IfNode");
    IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4248
    Node *iftrue = _gvn.transform(new (C) IfTrueNode(opt_ifisnan));
D
duke 已提交
4249 4250 4251 4252 4253 4254 4255 4256 4257

    set_control(iftrue);

    static const jlong nan_bits = CONST64(0x7ff8000000000000);
    Node *slow_result = longcon(nan_bits); // return NaN
    phi->init_req(1, _gvn.transform( slow_result ));
    r->init_req(1, iftrue);

    // Else fall through
4258
    Node *iffalse = _gvn.transform(new (C) IfFalseNode(opt_ifisnan));
D
duke 已提交
4259 4260
    set_control(iffalse);

4261
    phi->init_req(2, _gvn.transform(new (C) MoveD2LNode(arg)));
D
duke 已提交
4262 4263 4264 4265 4266 4267 4268
    r->init_req(2, iffalse);

    // Post merge
    set_control(_gvn.transform(r));
    record_for_igvn(r);

    C->set_has_split_ifs(true); // Has chance for split-if optimization
4269 4270
    result = phi;
    assert(result->bottom_type()->isa_long(), "must be");
D
duke 已提交
4271 4272 4273 4274 4275
    break;
  }

  case vmIntrinsics::_floatToIntBits: {
    // two paths (plus control) merge in a wood
4276 4277
    RegionNode *r = new (C) RegionNode(3);
    Node *phi = new (C) PhiNode(r, TypeInt::INT);
D
duke 已提交
4278

4279
    Node *cmpisnan = _gvn.transform(new (C) CmpFNode(arg, arg));
D
duke 已提交
4280
    // Build the boolean node
4281
    Node *bolisnan = _gvn.transform(new (C) BoolNode(cmpisnan, BoolTest::ne));
D
duke 已提交
4282 4283 4284 4285 4286 4287 4288

    // Branch either way.
    // NaN case is less traveled, which makes all the difference.
    IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
    Node *opt_isnan = _gvn.transform(ifisnan);
    assert( opt_isnan->is_If(), "Expect an IfNode");
    IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4289
    Node *iftrue = _gvn.transform(new (C) IfTrueNode(opt_ifisnan));
D
duke 已提交
4290 4291 4292 4293 4294 4295 4296 4297 4298

    set_control(iftrue);

    static const jint nan_bits = 0x7fc00000;
    Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
    phi->init_req(1, _gvn.transform( slow_result ));
    r->init_req(1, iftrue);

    // Else fall through
4299
    Node *iffalse = _gvn.transform(new (C) IfFalseNode(opt_ifisnan));
D
duke 已提交
4300 4301
    set_control(iffalse);

4302
    phi->init_req(2, _gvn.transform(new (C) MoveF2INode(arg)));
D
duke 已提交
4303 4304 4305 4306 4307 4308 4309
    r->init_req(2, iffalse);

    // Post merge
    set_control(_gvn.transform(r));
    record_for_igvn(r);

    C->set_has_split_ifs(true); // Has chance for split-if optimization
4310 4311
    result = phi;
    assert(result->bottom_type()->isa_int(), "must be");
D
duke 已提交
4312 4313 4314 4315
    break;
  }

  default:
4316 4317
    fatal_unexpected_iid(id);
    break;
D
duke 已提交
4318
  }
4319
  set_result(_gvn.transform(result));
D
duke 已提交
4320 4321 4322 4323 4324 4325 4326 4327 4328 4329
  return true;
}

#ifdef _LP64
#define XTOP ,top() /*additional argument*/
#else  //_LP64
#define XTOP        /*no additional argument*/
#endif //_LP64

//----------------------inline_unsafe_copyMemory-------------------------
4330
// public native void sun.misc.Unsafe.copyMemory(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
D
duke 已提交
4331 4332
bool LibraryCallKit::inline_unsafe_copyMemory() {
  if (callee()->is_static())  return false;  // caller must have the capability!
4333
  null_check_receiver();  // null-check receiver
D
duke 已提交
4334 4335 4336 4337
  if (stopped())  return true;

  C->set_has_unsafe_access(true);  // Mark eventual nmethod as "unsafe".

4338 4339 4340 4341 4342
  Node* src_ptr =         argument(1);   // type: oop
  Node* src_off = ConvL2X(argument(2));  // type: long
  Node* dst_ptr =         argument(4);   // type: oop
  Node* dst_off = ConvL2X(argument(5));  // type: long
  Node* size    = ConvL2X(argument(7));  // type: long
D
duke 已提交
4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367

  assert(Unsafe_field_offset_to_byte_offset(11) == 11,
         "fieldOffset must be byte-scaled");

  Node* src = make_unsafe_address(src_ptr, src_off);
  Node* dst = make_unsafe_address(dst_ptr, dst_off);

  // Conservatively insert a memory barrier on all memory slices.
  // Do not let writes of the copy source or destination float below the copy.
  insert_mem_bar(Op_MemBarCPUOrder);

  // Call it.  Note that the length argument is not scaled.
  make_runtime_call(RC_LEAF|RC_NO_FP,
                    OptoRuntime::fast_arraycopy_Type(),
                    StubRoutines::unsafe_arraycopy(),
                    "unsafe_arraycopy",
                    TypeRawPtr::BOTTOM,
                    src, dst, size XTOP);

  // Do not let reads of the copy destination float above the copy.
  insert_mem_bar(Op_MemBarCPUOrder);

  return true;
}

4368 4369 4370 4371 4372 4373 4374
//------------------------clone_coping-----------------------------------
// Helper function for inline_native_clone.
void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark) {
  assert(obj_size != NULL, "");
  Node* raw_obj = alloc_obj->in(1);
  assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");

4375
  AllocateNode* alloc = NULL;
4376 4377 4378
  if (ReduceBulkZeroing) {
    // We will be completely responsible for initializing this object -
    // mark Initialize node as complete.
4379
    alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
4380 4381
    // The object was just allocated - there should be no any stores!
    guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
4382 4383 4384 4385
    // Mark as complete_with_arraycopy so that on AllocateNode
    // expansion, we know this AllocateNode is initialized by an array
    // copy and a StoreStore barrier exists after the array copy.
    alloc->initialization()->set_complete_with_arraycopy();
4386 4387 4388 4389 4390
  }

  // Copy the fastest available way.
  // TODO: generate fields copies for small objects instead.
  Node* src  = obj;
4391
  Node* dest = alloc_obj;
4392 4393 4394 4395 4396 4397 4398 4399
  Node* size = _gvn.transform(obj_size);

  // Exclude the header but include array length to copy by 8 bytes words.
  // Can't use base_offset_in_bytes(bt) since basic type is unknown.
  int base_off = is_array ? arrayOopDesc::length_offset_in_bytes() :
                            instanceOopDesc::base_offset_in_bytes();
  // base_off:
  // 8  - 32-bit VM
4400 4401
  // 12 - 64-bit VM, compressed klass
  // 16 - 64-bit VM, normal klass
4402
  if (base_off % BytesPerLong != 0) {
4403
    assert(UseCompressedClassPointers, "");
4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417
    if (is_array) {
      // Exclude length to copy by 8 bytes words.
      base_off += sizeof(int);
    } else {
      // Include klass to copy by 8 bytes words.
      base_off = instanceOopDesc::klass_offset_in_bytes();
    }
    assert(base_off % BytesPerLong == 0, "expect 8 bytes alignment");
  }
  src  = basic_plus_adr(src,  base_off);
  dest = basic_plus_adr(dest, base_off);

  // Compute the length also, if needed:
  Node* countx = size;
4418 4419
  countx = _gvn.transform(new (C) SubXNode(countx, MakeConX(base_off)));
  countx = _gvn.transform(new (C) URShiftXNode(countx, intcon(LogBytesPerLong) ));
4420 4421 4422 4423

  const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
  bool disjoint_bases = true;
  generate_unchecked_arraycopy(raw_adr_type, T_LONG, disjoint_bases,
4424 4425
                               src, NULL, dest, NULL, countx,
                               /*dest_uninitialized*/true);
4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437

  // If necessary, emit some card marks afterwards.  (Non-arrays only.)
  if (card_mark) {
    assert(!is_array, "");
    // Put in store barrier for any and all oops we are sticking
    // into this object.  (We could avoid this if we could prove
    // that the object type contains no oop fields at all.)
    Node* no_particular_value = NULL;
    Node* no_particular_field = NULL;
    int raw_adr_idx = Compile::AliasIdxRaw;
    post_barrier(control(),
                 memory(raw_adr_type),
4438
                 alloc_obj,
4439 4440 4441 4442 4443 4444 4445
                 no_particular_field,
                 raw_adr_idx,
                 no_particular_value,
                 T_OBJECT,
                 false);
  }

4446
  // Do not let reads from the cloned object float above the arraycopy.
4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458
  if (alloc != NULL) {
    // Do not let stores that initialize this object be reordered with
    // a subsequent store that would make this object accessible by
    // other threads.
    // Record what AllocateNode this StoreStore protects so that
    // escape analysis can go from the MemBarStoreStoreNode to the
    // AllocateNode and eliminate the MemBarStoreStoreNode if possible
    // based on the escape status of the AllocateNode.
    insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
  } else {
    insert_mem_bar(Op_MemBarCPUOrder);
  }
4459
}
D
duke 已提交
4460 4461

//------------------------inline_native_clone----------------------------
4462 4463
// protected native Object java.lang.Object.clone();
//
D
duke 已提交
4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479
// Here are the simple edge cases:
//  null receiver => normal trap
//  virtual and clone was overridden => slow path to out-of-line clone
//  not cloneable or finalizer => slow path to out-of-line Object.clone
//
// The general case has two steps, allocation and copying.
// Allocation has two cases, and uses GraphKit::new_instance or new_array.
//
// Copying also has two cases, oop arrays and everything else.
// Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
// Everything else uses the tight inline loop supplied by CopyArrayNode.
//
// These steps fold up nicely if and when the cloned object's klass
// can be sharply typed as an object array, a type array, or an instance.
//
bool LibraryCallKit::inline_native_clone(bool is_virtual) {
4480
  PhiNode* result_val;
D
duke 已提交
4481

4482 4483
  // Set the reexecute bit for the interpreter to reexecute
  // the bytecode that invokes Object.clone if deoptimization happens.
4484 4485 4486
  { PreserveReexecuteState preexecs(this);
    jvms()->set_should_reexecute(true);

4487
    Node* obj = null_check_receiver();
4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507
    if (stopped())  return true;

    Node* obj_klass = load_object_klass(obj);
    const TypeKlassPtr* tklass = _gvn.type(obj_klass)->isa_klassptr();
    const TypeOopPtr*   toop   = ((tklass != NULL)
                                ? tklass->as_instance_type()
                                : TypeInstPtr::NOTNULL);

    // Conservatively insert a memory barrier on all memory slices.
    // Do not let writes into the original float below the clone.
    insert_mem_bar(Op_MemBarCPUOrder);

    // paths into result_reg:
    enum {
      _slow_path = 1,     // out-of-line call to clone method (virtual or not)
      _objArray_path,     // plain array allocation, plus arrayof_oop_arraycopy
      _array_path,        // plain array allocation, plus arrayof_long_arraycopy
      _instance_path,     // plain instance allocation, plus arrayof_long_arraycopy
      PATH_LIMIT
    };
4508 4509 4510 4511 4512 4513
    RegionNode* result_reg = new(C) RegionNode(PATH_LIMIT);
    result_val             = new(C) PhiNode(result_reg,
                                            TypeInstPtr::NOTNULL);
    PhiNode*    result_i_o = new(C) PhiNode(result_reg, Type::ABIO);
    PhiNode*    result_mem = new(C) PhiNode(result_reg, Type::MEMORY,
                                            TypePtr::BOTTOM);
4514 4515 4516 4517 4518
    record_for_igvn(result_reg);

    const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
    int raw_adr_idx = Compile::AliasIdxRaw;

4519 4520 4521 4522 4523 4524 4525
    Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)NULL);
    if (array_ctl != NULL) {
      // It's an array.
      PreserveJVMState pjvms(this);
      set_control(array_ctl);
      Node* obj_length = load_array_length(obj);
      Node* obj_size  = NULL;
4526
      Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size);  // no arguments to push
4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548

      if (!use_ReduceInitialCardMarks()) {
        // If it is an oop array, it requires very special treatment,
        // because card marking is required on each card of the array.
        Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)NULL);
        if (is_obja != NULL) {
          PreserveJVMState pjvms2(this);
          set_control(is_obja);
          // Generate a direct call to the right arraycopy function(s).
          bool disjoint_bases = true;
          bool length_never_negative = true;
          generate_arraycopy(TypeAryPtr::OOPS, T_OBJECT,
                             obj, intcon(0), alloc_obj, intcon(0),
                             obj_length,
                             disjoint_bases, length_never_negative);
          result_reg->init_req(_objArray_path, control());
          result_val->init_req(_objArray_path, alloc_obj);
          result_i_o ->set_req(_objArray_path, i_o());
          result_mem ->set_req(_objArray_path, reset_memory());
        }
      }
      // Otherwise, there are no card marks to worry about.
4549 4550 4551 4552 4553 4554
      // (We can dispense with card marks if we know the allocation
      //  comes out of eden (TLAB)...  In fact, ReduceInitialCardMarks
      //  causes the non-eden paths to take compensating steps to
      //  simulate a fresh allocation, so that no further
      //  card marks are required in compiled code to initialize
      //  the object.)
4555 4556 4557 4558 4559 4560 4561 4562 4563

      if (!stopped()) {
        copy_to_clone(obj, alloc_obj, obj_size, true, false);

        // Present the results of the copy.
        result_reg->init_req(_array_path, control());
        result_val->init_req(_array_path, alloc_obj);
        result_i_o ->set_req(_array_path, i_o());
        result_mem ->set_req(_array_path, reset_memory());
D
duke 已提交
4564 4565
      }
    }
4566

4567 4568
    // We only go to the instance fast case code if we pass a number of guards.
    // The paths which do not pass are accumulated in the slow_region.
4569
    RegionNode* slow_region = new (C) RegionNode(1);
4570
    record_for_igvn(slow_region);
4571
    if (!stopped()) {
4572 4573 4574 4575 4576 4577 4578 4579 4580 4581
      // It's an instance (we did array above).  Make the slow-path tests.
      // If this is a virtual call, we generate a funny guard.  We grab
      // the vtable entry corresponding to clone() from the target object.
      // If the target method which we are calling happens to be the
      // Object clone() method, we pass the guard.  We do not need this
      // guard for non-virtual calls; the caller is known to be the native
      // Object clone().
      if (is_virtual) {
        generate_virtual_guard(obj_klass, slow_region);
      }
4582

4583 4584 4585 4586 4587 4588 4589 4590 4591
      // The object must be cloneable and must not have a finalizer.
      // Both of these conditions may be checked in a single test.
      // We could optimize the cloneable test further, but we don't care.
      generate_access_flags_guard(obj_klass,
                                  // Test both conditions:
                                  JVM_ACC_IS_CLONEABLE | JVM_ACC_HAS_FINALIZER,
                                  // Must be cloneable but not finalizer:
                                  JVM_ACC_IS_CLONEABLE,
                                  slow_region);
4592
    }
D
duke 已提交
4593

4594 4595 4596 4597
    if (!stopped()) {
      // It's an instance, and it passed the slow-path tests.
      PreserveJVMState pjvms(this);
      Node* obj_size  = NULL;
4598
      Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size);
4599 4600 4601 4602 4603 4604 4605 4606

      copy_to_clone(obj, alloc_obj, obj_size, false, !use_ReduceInitialCardMarks());

      // Present the results of the slow call.
      result_reg->init_req(_instance_path, control());
      result_val->init_req(_instance_path, alloc_obj);
      result_i_o ->set_req(_instance_path, i_o());
      result_mem ->set_req(_instance_path, reset_memory());
D
duke 已提交
4607 4608
    }

4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620
    // Generate code for the slow case.  We make a call to clone().
    set_control(_gvn.transform(slow_region));
    if (!stopped()) {
      PreserveJVMState pjvms(this);
      CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
      Node* slow_result = set_results_for_java_call(slow_call);
      // this->control() comes from set_results_for_java_call
      result_reg->init_req(_slow_path, control());
      result_val->init_req(_slow_path, slow_result);
      result_i_o ->set_req(_slow_path, i_o());
      result_mem ->set_req(_slow_path, reset_memory());
    }
D
duke 已提交
4621

4622
    // Return the combined state.
4623 4624 4625
    set_control(    _gvn.transform(result_reg));
    set_i_o(        _gvn.transform(result_i_o));
    set_all_memory( _gvn.transform(result_mem));
4626
  } // original reexecute is set back here
D
duke 已提交
4627

4628
  set_result(_gvn.transform(result_val));
D
duke 已提交
4629 4630 4631 4632 4633 4634 4635 4636
  return true;
}

//------------------------------basictype2arraycopy----------------------------
address LibraryCallKit::basictype2arraycopy(BasicType t,
                                            Node* src_offset,
                                            Node* dest_offset,
                                            bool disjoint_bases,
4637 4638
                                            const char* &name,
                                            bool dest_uninitialized) {
D
duke 已提交
4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654
  const TypeInt* src_offset_inttype  = gvn().find_int_type(src_offset);;
  const TypeInt* dest_offset_inttype = gvn().find_int_type(dest_offset);;

  bool aligned = false;
  bool disjoint = disjoint_bases;

  // if the offsets are the same, we can treat the memory regions as
  // disjoint, because either the memory regions are in different arrays,
  // or they are identical (which we can treat as disjoint.)  We can also
  // treat a copy with a destination index  less that the source index
  // as disjoint since a low->high copy will work correctly in this case.
  if (src_offset_inttype != NULL && src_offset_inttype->is_con() &&
      dest_offset_inttype != NULL && dest_offset_inttype->is_con()) {
    // both indices are constants
    int s_offs = src_offset_inttype->get_con();
    int d_offs = dest_offset_inttype->get_con();
4655
    int element_size = type2aelembytes(t);
D
duke 已提交
4656 4657 4658 4659 4660 4661 4662 4663
    aligned = ((arrayOopDesc::base_offset_in_bytes(t) + s_offs * element_size) % HeapWordSize == 0) &&
              ((arrayOopDesc::base_offset_in_bytes(t) + d_offs * element_size) % HeapWordSize == 0);
    if (s_offs >= d_offs)  disjoint = true;
  } else if (src_offset == dest_offset && src_offset != NULL) {
    // This can occur if the offsets are identical non-constants.
    disjoint = true;
  }

4664
  return StubRoutines::select_arraycopy_function(t, aligned, disjoint, name, dest_uninitialized);
D
duke 已提交
4665 4666 4667 4668
}


//------------------------------inline_arraycopy-----------------------
4669 4670 4671
// public static native void java.lang.System.arraycopy(Object src,  int  srcPos,
//                                                      Object dest, int destPos,
//                                                      int length);
D
duke 已提交
4672
bool LibraryCallKit::inline_arraycopy() {
4673 4674 4675 4676 4677 4678
  // Get the arguments.
  Node* src         = argument(0);  // type: oop
  Node* src_offset  = argument(1);  // type: int
  Node* dest        = argument(2);  // type: oop
  Node* dest_offset = argument(3);  // type: int
  Node* length      = argument(4);  // type: int
D
duke 已提交
4679 4680 4681 4682 4683 4684

  // Compile time checks.  If any of these checks cannot be verified at compile time,
  // we do not make a fast path for this call.  Instead, we let the call remain as it
  // is.  The checks we choose to mandate at compile time are:
  //
  // (1) src and dest are arrays.
4685
  const Type* src_type  = src->Value(&_gvn);
D
duke 已提交
4686
  const Type* dest_type = dest->Value(&_gvn);
4687
  const TypeAryPtr* top_src  = src_type->isa_aryptr();
D
duke 已提交
4688
  const TypeAryPtr* top_dest = dest_type->isa_aryptr();
4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744

  // Do we have the type of src?
  bool has_src = (top_src != NULL && top_src->klass() != NULL);
  // Do we have the type of dest?
  bool has_dest = (top_dest != NULL && top_dest->klass() != NULL);
  // Is the type for src from speculation?
  bool src_spec = false;
  // Is the type for dest from speculation?
  bool dest_spec = false;

  if (!has_src || !has_dest) {
    // We don't have sufficient type information, let's see if
    // speculative types can help. We need to have types for both src
    // and dest so that it pays off.

    // Do we already have or could we have type information for src
    bool could_have_src = has_src;
    // Do we already have or could we have type information for dest
    bool could_have_dest = has_dest;

    ciKlass* src_k = NULL;
    if (!has_src) {
      src_k = src_type->speculative_type();
      if (src_k != NULL && src_k->is_array_klass()) {
        could_have_src = true;
      }
    }

    ciKlass* dest_k = NULL;
    if (!has_dest) {
      dest_k = dest_type->speculative_type();
      if (dest_k != NULL && dest_k->is_array_klass()) {
        could_have_dest = true;
      }
    }

    if (could_have_src && could_have_dest) {
      // This is going to pay off so emit the required guards
      if (!has_src) {
        src = maybe_cast_profiled_obj(src, src_k);
        src_type  = _gvn.type(src);
        top_src  = src_type->isa_aryptr();
        has_src = (top_src != NULL && top_src->klass() != NULL);
        src_spec = true;
      }
      if (!has_dest) {
        dest = maybe_cast_profiled_obj(dest, dest_k);
        dest_type  = _gvn.type(dest);
        top_dest  = dest_type->isa_aryptr();
        has_dest = (top_dest != NULL && top_dest->klass() != NULL);
        dest_spec = true;
      }
    }
  }

  if (!has_src || !has_dest) {
D
duke 已提交
4745 4746 4747 4748 4749 4750
    // Conservatively insert a memory barrier on all memory slices.
    // Do not let writes into the source float below the arraycopy.
    insert_mem_bar(Op_MemBarCPUOrder);

    // Call StubRoutines::generic_arraycopy stub.
    generate_arraycopy(TypeRawPtr::BOTTOM, T_CONFLICT,
4751
                       src, src_offset, dest, dest_offset, length);
D
duke 已提交
4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773

    // Do not let reads from the destination float above the arraycopy.
    // Since we cannot type the arrays, we don't know which slices
    // might be affected.  We could restrict this barrier only to those
    // memory slices which pertain to array elements--but don't bother.
    if (!InsertMemBarAfterArraycopy)
      // (If InsertMemBarAfterArraycopy, there is already one in place.)
      insert_mem_bar(Op_MemBarCPUOrder);
    return true;
  }

  // (2) src and dest arrays must have elements of the same BasicType
  // Figure out the size and type of the elements we will be copying.
  BasicType src_elem  =  top_src->klass()->as_array_klass()->element_type()->basic_type();
  BasicType dest_elem = top_dest->klass()->as_array_klass()->element_type()->basic_type();
  if (src_elem  == T_ARRAY)  src_elem  = T_OBJECT;
  if (dest_elem == T_ARRAY)  dest_elem = T_OBJECT;

  if (src_elem != dest_elem || dest_elem == T_VOID) {
    // The component types are not the same or are not recognized.  Punt.
    // (But, avoid the native method wrapper to JVM_ArrayCopy.)
    generate_slow_arraycopy(TypePtr::BOTTOM,
4774 4775
                            src, src_offset, dest, dest_offset, length,
                            /*dest_uninitialized*/false);
D
duke 已提交
4776 4777 4778
    return true;
  }

4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812
  if (src_elem == T_OBJECT) {
    // If both arrays are object arrays then having the exact types
    // for both will remove the need for a subtype check at runtime
    // before the call and may make it possible to pick a faster copy
    // routine (without a subtype check on every element)
    // Do we have the exact type of src?
    bool could_have_src = src_spec;
    // Do we have the exact type of dest?
    bool could_have_dest = dest_spec;
    ciKlass* src_k = top_src->klass();
    ciKlass* dest_k = top_dest->klass();
    if (!src_spec) {
      src_k = src_type->speculative_type();
      if (src_k != NULL && src_k->is_array_klass()) {
          could_have_src = true;
      }
    }
    if (!dest_spec) {
      dest_k = dest_type->speculative_type();
      if (dest_k != NULL && dest_k->is_array_klass()) {
        could_have_dest = true;
      }
    }
    if (could_have_src && could_have_dest) {
      // If we can have both exact types, emit the missing guards
      if (could_have_src && !src_spec) {
        src = maybe_cast_profiled_obj(src, src_k);
      }
      if (could_have_dest && !dest_spec) {
        dest = maybe_cast_profiled_obj(dest, dest_k);
      }
    }
  }

D
duke 已提交
4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825
  //---------------------------------------------------------------------------
  // We will make a fast path for this call to arraycopy.

  // We have the following tests left to perform:
  //
  // (3) src and dest must not be null.
  // (4) src_offset must not be negative.
  // (5) dest_offset must not be negative.
  // (6) length must not be negative.
  // (7) src_offset + length must not exceed length of src.
  // (8) dest_offset + length must not exceed length of dest.
  // (9) each element of an oop array must be assignable

4826
  RegionNode* slow_region = new (C) RegionNode(1);
D
duke 已提交
4827 4828 4829
  record_for_igvn(slow_region);

  // (3) operands must not be null
4830
  // We currently perform our null checks with the null_check routine.
D
duke 已提交
4831 4832 4833 4834
  // This means that the null exceptions will be reported in the caller
  // rather than (correctly) reported inside of the native arraycopy call.
  // This should be corrected, given time.  We do our null check with the
  // stack pointer restored.
4835 4836
  src  = null_check(src,  T_ARRAY);
  dest = null_check(dest, T_ARRAY);
D
duke 已提交
4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863

  // (4) src_offset must not be negative.
  generate_negative_guard(src_offset, slow_region);

  // (5) dest_offset must not be negative.
  generate_negative_guard(dest_offset, slow_region);

  // (6) length must not be negative (moved to generate_arraycopy()).
  // generate_negative_guard(length, slow_region);

  // (7) src_offset + length must not exceed length of src.
  generate_limit_guard(src_offset, length,
                       load_array_length(src),
                       slow_region);

  // (8) dest_offset + length must not exceed length of dest.
  generate_limit_guard(dest_offset, length,
                       load_array_length(dest),
                       slow_region);

  // (9) each element of an oop array must be assignable
  // The generate_arraycopy subroutine checks this.

  // This is where the memory effects are placed:
  const TypePtr* adr_type = TypeAryPtr::get_array_body_type(dest_elem);
  generate_arraycopy(adr_type, dest_elem,
                     src, src_offset, dest, dest_offset, length,
4864
                     false, false, slow_region);
D
duke 已提交
4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913

  return true;
}

//-----------------------------generate_arraycopy----------------------
// Generate an optimized call to arraycopy.
// Caller must guard against non-arrays.
// Caller must determine a common array basic-type for both arrays.
// Caller must validate offsets against array bounds.
// The slow_region has already collected guard failure paths
// (such as out of bounds length or non-conformable array types).
// The generated code has this shape, in general:
//
//     if (length == 0)  return   // via zero_path
//     slowval = -1
//     if (types unknown) {
//       slowval = call generic copy loop
//       if (slowval == 0)  return  // via checked_path
//     } else if (indexes in bounds) {
//       if ((is object array) && !(array type check)) {
//         slowval = call checked copy loop
//         if (slowval == 0)  return  // via checked_path
//       } else {
//         call bulk copy loop
//         return  // via fast_path
//       }
//     }
//     // adjust params for remaining work:
//     if (slowval != -1) {
//       n = -1^slowval; src_offset += n; dest_offset += n; length -= n
//     }
//   slow_region:
//     call slow arraycopy(src, src_offset, dest, dest_offset, length)
//     return  // via slow_call_path
//
// This routine is used from several intrinsics:  System.arraycopy,
// Object.clone (the array subcase), and Arrays.copyOf[Range].
//
void
LibraryCallKit::generate_arraycopy(const TypePtr* adr_type,
                                   BasicType basic_elem_type,
                                   Node* src,  Node* src_offset,
                                   Node* dest, Node* dest_offset,
                                   Node* copy_length,
                                   bool disjoint_bases,
                                   bool length_never_negative,
                                   RegionNode* slow_region) {

  if (slow_region == NULL) {
4914
    slow_region = new(C) RegionNode(1);
D
duke 已提交
4915 4916 4917 4918 4919
    record_for_igvn(slow_region);
  }

  Node* original_dest      = dest;
  AllocateArrayNode* alloc = NULL;  // used for zeroing, if needed
4920
  bool  dest_uninitialized = false;
D
duke 已提交
4921 4922 4923 4924 4925 4926 4927 4928 4929

  // See if this is the initialization of a newly-allocated array.
  // If so, we will take responsibility here for initializing it to zero.
  // (Note:  Because tightly_coupled_allocation performs checks on the
  // out-edges of the dest, we need to avoid making derived pointers
  // from it until we have checked its uses.)
  if (ReduceBulkZeroing
      && !ZeroTLAB              // pointless if already zeroed
      && basic_elem_type != T_CONFLICT // avoid corner case
K
kvn 已提交
4930
      && !src->eqv_uncast(dest)
D
duke 已提交
4931 4932
      && ((alloc = tightly_coupled_allocation(dest, slow_region))
          != NULL)
4933
      && _gvn.find_int_con(alloc->in(AllocateNode::ALength), 1) > 0
D
duke 已提交
4934 4935 4936 4937
      && alloc->maybe_set_complete(&_gvn)) {
    // "You break it, you buy it."
    InitializeNode* init = alloc->initialization();
    assert(init->is_complete(), "we just did this");
4938
    init->set_complete_with_arraycopy();
4939
    assert(dest->is_CheckCastPP(), "sanity");
D
duke 已提交
4940 4941 4942 4943
    assert(dest->in(0)->in(0) == init, "dest pinned");
    adr_type = TypeRawPtr::BOTTOM;  // all initializations are into raw memory
    // From this point on, every exit path is responsible for
    // initializing any non-copied parts of the object to zero.
4944 4945 4946
    // Also, if this flag is set we make sure that arraycopy interacts properly
    // with G1, eliding pre-barriers. See CR 6627983.
    dest_uninitialized = true;
D
duke 已提交
4947 4948 4949 4950
  } else {
    // No zeroing elimination here.
    alloc             = NULL;
    //original_dest   = dest;
4951
    //dest_uninitialized = false;
D
duke 已提交
4952 4953 4954 4955 4956 4957 4958 4959 4960 4961
  }

  // Results are placed here:
  enum { fast_path        = 1,  // normal void-returning assembly stub
         checked_path     = 2,  // special assembly stub with cleanup
         slow_call_path   = 3,  // something went wrong; call the VM
         zero_path        = 4,  // bypass when length of copy is zero
         bcopy_path       = 5,  // copy primitive array by 64-bit blocks
         PATH_LIMIT       = 6
  };
4962 4963 4964
  RegionNode* result_region = new(C) RegionNode(PATH_LIMIT);
  PhiNode*    result_i_o    = new(C) PhiNode(result_region, Type::ABIO);
  PhiNode*    result_memory = new(C) PhiNode(result_region, Type::MEMORY, adr_type);
D
duke 已提交
4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982
  record_for_igvn(result_region);
  _gvn.set_type_bottom(result_i_o);
  _gvn.set_type_bottom(result_memory);
  assert(adr_type != TypePtr::BOTTOM, "must be RawMem or a T[] slice");

  // The slow_control path:
  Node* slow_control;
  Node* slow_i_o = i_o();
  Node* slow_mem = memory(adr_type);
  debug_only(slow_control = (Node*) badAddress);

  // Checked control path:
  Node* checked_control = top();
  Node* checked_mem     = NULL;
  Node* checked_i_o     = NULL;
  Node* checked_value   = NULL;

  if (basic_elem_type == T_CONFLICT) {
4983
    assert(!dest_uninitialized, "");
D
duke 已提交
4984 4985
    Node* cv = generate_generic_arraycopy(adr_type,
                                          src, src_offset, dest, dest_offset,
4986
                                          copy_length, dest_uninitialized);
D
duke 已提交
4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004
    if (cv == NULL)  cv = intcon(-1);  // failure (no stub available)
    checked_control = control();
    checked_i_o     = i_o();
    checked_mem     = memory(adr_type);
    checked_value   = cv;
    set_control(top());         // no fast path
  }

  Node* not_pos = generate_nonpositive_guard(copy_length, length_never_negative);
  if (not_pos != NULL) {
    PreserveJVMState pjvms(this);
    set_control(not_pos);

    // (6) length must not be negative.
    if (!length_never_negative) {
      generate_negative_guard(copy_length, slow_region);
    }

5005
    // copy_length is 0.
5006
    if (!stopped() && dest_uninitialized) {
D
duke 已提交
5007
      Node* dest_length = alloc->in(AllocateNode::ALength);
K
kvn 已提交
5008
      if (copy_length->eqv_uncast(dest_length)
D
duke 已提交
5009
          || _gvn.find_int_con(dest_length, 1) <= 0) {
5010
        // There is no zeroing to do. No need for a secondary raw memory barrier.
D
duke 已提交
5011 5012 5013 5014 5015
      } else {
        // Clear the whole thing since there are no source elements to copy.
        generate_clear_array(adr_type, dest, basic_elem_type,
                             intcon(0), NULL,
                             alloc->in(AllocateNode::AllocSize));
5016 5017 5018 5019 5020 5021 5022
        // Use a secondary InitializeNode as raw memory barrier.
        // Currently it is needed only on this path since other
        // paths have stub or runtime calls as raw memory barriers.
        InitializeNode* init = insert_mem_bar_volatile(Op_Initialize,
                                                       Compile::AliasIdxRaw,
                                                       top())->as_Initialize();
        init->set_complete(&_gvn);  // (there is no corresponding AllocateNode)
D
duke 已提交
5023 5024 5025 5026 5027 5028 5029 5030 5031
      }
    }

    // Present the results of the fast call.
    result_region->init_req(zero_path, control());
    result_i_o   ->init_req(zero_path, i_o());
    result_memory->init_req(zero_path, memory(adr_type));
  }

5032
  if (!stopped() && dest_uninitialized) {
D
duke 已提交
5033 5034 5035 5036 5037
    // We have to initialize the *uncopied* part of the array to zero.
    // The copy destination is the slice dest[off..off+len].  The other slices
    // are dest_head = dest[0..off] and dest_tail = dest[off+len..dest.length].
    Node* dest_size   = alloc->in(AllocateNode::AllocSize);
    Node* dest_length = alloc->in(AllocateNode::ALength);
5038 5039
    Node* dest_tail   = _gvn.transform(new(C) AddINode(dest_offset,
                                                          copy_length));
D
duke 已提交
5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053

    // If there is a head section that needs zeroing, do it now.
    if (find_int_con(dest_offset, -1) != 0) {
      generate_clear_array(adr_type, dest, basic_elem_type,
                           intcon(0), dest_offset,
                           NULL);
    }

    // Next, perform a dynamic check on the tail length.
    // It is often zero, and we can win big if we prove this.
    // There are two wins:  Avoid generating the ClearArray
    // with its attendant messy index arithmetic, and upgrade
    // the copy to a more hardware-friendly word size of 64 bits.
    Node* tail_ctl = NULL;
K
kvn 已提交
5054
    if (!stopped() && !dest_tail->eqv_uncast(dest_length)) {
5055 5056
      Node* cmp_lt   = _gvn.transform(new(C) CmpINode(dest_tail, dest_length));
      Node* bol_lt   = _gvn.transform(new(C) BoolNode(cmp_lt, BoolTest::lt));
D
duke 已提交
5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067
      tail_ctl = generate_slow_guard(bol_lt, NULL);
      assert(tail_ctl != NULL || !stopped(), "must be an outcome");
    }

    // At this point, let's assume there is no tail.
    if (!stopped() && alloc != NULL && basic_elem_type != T_OBJECT) {
      // There is no tail.  Try an upgrade to a 64-bit copy.
      bool didit = false;
      { PreserveJVMState pjvms(this);
        didit = generate_block_arraycopy(adr_type, basic_elem_type, alloc,
                                         src, src_offset, dest, dest_offset,
5068
                                         dest_size, dest_uninitialized);
D
duke 已提交
5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089
        if (didit) {
          // Present the results of the block-copying fast call.
          result_region->init_req(bcopy_path, control());
          result_i_o   ->init_req(bcopy_path, i_o());
          result_memory->init_req(bcopy_path, memory(adr_type));
        }
      }
      if (didit)
        set_control(top());     // no regular fast path
    }

    // Clear the tail, if any.
    if (tail_ctl != NULL) {
      Node* notail_ctl = stopped() ? NULL : control();
      set_control(tail_ctl);
      if (notail_ctl == NULL) {
        generate_clear_array(adr_type, dest, basic_elem_type,
                             dest_tail, NULL,
                             dest_size);
      } else {
        // Make a local merge.
5090 5091
        Node* done_ctl = new(C) RegionNode(3);
        Node* done_mem = new(C) PhiNode(done_ctl, Type::MEMORY, adr_type);
D
duke 已提交
5092 5093 5094 5095 5096 5097 5098
        done_ctl->init_req(1, notail_ctl);
        done_mem->init_req(1, memory(adr_type));
        generate_clear_array(adr_type, dest, basic_elem_type,
                             dest_tail, NULL,
                             dest_size);
        done_ctl->init_req(2, control());
        done_mem->init_req(2, memory(adr_type));
5099
        set_control( _gvn.transform(done_ctl));
D
duke 已提交
5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115
        set_memory(  _gvn.transform(done_mem), adr_type );
      }
    }
  }

  BasicType copy_type = basic_elem_type;
  assert(basic_elem_type != T_ARRAY, "caller must fix this");
  if (!stopped() && copy_type == T_OBJECT) {
    // If src and dest have compatible element types, we can copy bits.
    // Types S[] and D[] are compatible if D is a supertype of S.
    //
    // If they are not, we will use checked_oop_disjoint_arraycopy,
    // which performs a fast optimistic per-oop check, and backs off
    // further to JVM_ArrayCopy on the first per-oop check that fails.
    // (Actually, we don't move raw bits only; the GC requires card marks.)

5116
    // Get the Klass* for both src and dest
D
duke 已提交
5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136
    Node* src_klass  = load_object_klass(src);
    Node* dest_klass = load_object_klass(dest);

    // Generate the subtype check.
    // This might fold up statically, or then again it might not.
    //
    // Non-static example:  Copying List<String>.elements to a new String[].
    // The backing store for a List<String> is always an Object[],
    // but its elements are always type String, if the generic types
    // are correct at the source level.
    //
    // Test S[] against D[], not S against D, because (probably)
    // the secondary supertype cache is less busy for S[] than S.
    // This usually only matters when D is an interface.
    Node* not_subtype_ctrl = gen_subtype_check(src_klass, dest_klass);
    // Plug failing path into checked_oop_disjoint_arraycopy
    if (not_subtype_ctrl != top()) {
      PreserveJVMState pjvms(this);
      set_control(not_subtype_ctrl);
      // (At this point we can assume disjoint_bases, since types differ.)
5137
      int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset());
D
duke 已提交
5138
      Node* p1 = basic_plus_adr(dest_klass, ek_offset);
5139
      Node* n1 = LoadKlassNode::make(_gvn, immutable_memory(), p1, TypeRawPtr::BOTTOM);
D
duke 已提交
5140 5141 5142 5143
      Node* dest_elem_klass = _gvn.transform(n1);
      Node* cv = generate_checkcast_arraycopy(adr_type,
                                              dest_elem_klass,
                                              src, src_offset, dest, dest_offset,
5144
                                              ConvI2X(copy_length), dest_uninitialized);
D
duke 已提交
5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155
      if (cv == NULL)  cv = intcon(-1);  // failure (no stub available)
      checked_control = control();
      checked_i_o     = i_o();
      checked_mem     = memory(adr_type);
      checked_value   = cv;
    }
    // At this point we know we do not need type checks on oop stores.

    // Let's see if we need card marks:
    if (alloc != NULL && use_ReduceInitialCardMarks()) {
      // If we do not need card marks, copy using the jint or jlong stub.
5156
      copy_type = LP64_ONLY(UseCompressedOops ? T_INT : T_LONG) NOT_LP64(T_INT);
5157
      assert(type2aelembytes(basic_elem_type) == type2aelembytes(copy_type),
D
duke 已提交
5158 5159 5160 5161 5162 5163 5164 5165 5166
             "sizes agree");
    }
  }

  if (!stopped()) {
    // Generate the fast path, if possible.
    PreserveJVMState pjvms(this);
    generate_unchecked_arraycopy(adr_type, copy_type, disjoint_bases,
                                 src, src_offset, dest, dest_offset,
5167
                                 ConvI2X(copy_length), dest_uninitialized);
D
duke 已提交
5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178

    // Present the results of the fast call.
    result_region->init_req(fast_path, control());
    result_i_o   ->init_req(fast_path, i_o());
    result_memory->init_req(fast_path, memory(adr_type));
  }

  // Here are all the slow paths up to this point, in one bundle:
  slow_control = top();
  if (slow_region != NULL)
    slow_control = _gvn.transform(slow_region);
5179
  DEBUG_ONLY(slow_region = (RegionNode*)badAddress);
D
duke 已提交
5180 5181 5182 5183 5184 5185

  set_control(checked_control);
  if (!stopped()) {
    // Clean up after the checked call.
    // The returned value is either 0 or -1^K,
    // where K = number of partially transferred array elements.
5186 5187
    Node* cmp = _gvn.transform(new(C) CmpINode(checked_value, intcon(0)));
    Node* bol = _gvn.transform(new(C) BoolNode(cmp, BoolTest::eq));
D
duke 已提交
5188 5189 5190
    IfNode* iff = create_and_map_if(control(), bol, PROB_MAX, COUNT_UNKNOWN);

    // If it is 0, we are done, so transfer to the end.
5191
    Node* checks_done = _gvn.transform(new(C) IfTrueNode(iff));
D
duke 已提交
5192 5193 5194 5195 5196
    result_region->init_req(checked_path, checks_done);
    result_i_o   ->init_req(checked_path, checked_i_o);
    result_memory->init_req(checked_path, checked_mem);

    // If it is not zero, merge into the slow call.
5197
    set_control( _gvn.transform(new(C) IfFalseNode(iff) ));
5198 5199 5200
    RegionNode* slow_reg2 = new(C) RegionNode(3);
    PhiNode*    slow_i_o2 = new(C) PhiNode(slow_reg2, Type::ABIO);
    PhiNode*    slow_mem2 = new(C) PhiNode(slow_reg2, Type::MEMORY, adr_type);
D
duke 已提交
5201 5202 5203 5204 5205
    record_for_igvn(slow_reg2);
    slow_reg2  ->init_req(1, slow_control);
    slow_i_o2  ->init_req(1, slow_i_o);
    slow_mem2  ->init_req(1, slow_mem);
    slow_reg2  ->init_req(2, control());
5206 5207
    slow_i_o2  ->init_req(2, checked_i_o);
    slow_mem2  ->init_req(2, checked_mem);
D
duke 已提交
5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219

    slow_control = _gvn.transform(slow_reg2);
    slow_i_o     = _gvn.transform(slow_i_o2);
    slow_mem     = _gvn.transform(slow_mem2);

    if (alloc != NULL) {
      // We'll restart from the very beginning, after zeroing the whole thing.
      // This can cause double writes, but that's OK since dest is brand new.
      // So we ignore the low 31 bits of the value returned from the stub.
    } else {
      // We must continue the copy exactly where it failed, or else
      // another thread might see the wrong number of writes to dest.
5220
      Node* checked_offset = _gvn.transform(new(C) XorINode(checked_value, intcon(-1)));
5221
      Node* slow_offset    = new(C) PhiNode(slow_reg2, TypeInt::INT);
D
duke 已提交
5222 5223 5224 5225 5226
      slow_offset->init_req(1, intcon(0));
      slow_offset->init_req(2, checked_offset);
      slow_offset  = _gvn.transform(slow_offset);

      // Adjust the arguments by the conditionally incoming offset.
5227 5228 5229
      Node* src_off_plus  = _gvn.transform(new(C) AddINode(src_offset,  slow_offset));
      Node* dest_off_plus = _gvn.transform(new(C) AddINode(dest_offset, slow_offset));
      Node* length_minus  = _gvn.transform(new(C) SubINode(copy_length, slow_offset));
D
duke 已提交
5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245

      // Tweak the node variables to adjust the code produced below:
      src_offset  = src_off_plus;
      dest_offset = dest_off_plus;
      copy_length = length_minus;
    }
  }

  set_control(slow_control);
  if (!stopped()) {
    // Generate the slow path, if needed.
    PreserveJVMState pjvms(this);   // replace_in_map may trash the map

    set_memory(slow_mem, adr_type);
    set_i_o(slow_i_o);

5246
    if (dest_uninitialized) {
D
duke 已提交
5247 5248 5249 5250 5251 5252 5253
      generate_clear_array(adr_type, dest, basic_elem_type,
                           intcon(0), NULL,
                           alloc->in(AllocateNode::AllocSize));
    }

    generate_slow_arraycopy(adr_type,
                            src, src_offset, dest, dest_offset,
5254
                            copy_length, /*dest_uninitialized*/false);
D
duke 已提交
5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266 5267

    result_region->init_req(slow_call_path, control());
    result_i_o   ->init_req(slow_call_path, i_o());
    result_memory->init_req(slow_call_path, memory(adr_type));
  }

  // Remove unused edges.
  for (uint i = 1; i < result_region->req(); i++) {
    if (result_region->in(i) == NULL)
      result_region->init_req(i, top());
  }

  // Finished; return the combined state.
5268
  set_control( _gvn.transform(result_region));
D
duke 已提交
5269 5270 5271 5272
  set_i_o(     _gvn.transform(result_i_o)    );
  set_memory(  _gvn.transform(result_memory), adr_type );

  // The memory edges above are precise in order to model effects around
T
twisti 已提交
5273
  // array copies accurately to allow value numbering of field loads around
D
duke 已提交
5274 5275 5276 5277 5278 5279 5280 5281
  // arraycopy.  Such field loads, both before and after, are common in Java
  // collections and similar classes involving header/array data structures.
  //
  // But with low number of register or when some registers are used or killed
  // by arraycopy calls it causes registers spilling on stack. See 6544710.
  // The next memory barrier is added to avoid it. If the arraycopy can be
  // optimized away (which it can, sometimes) then we can manually remove
  // the membar also.
5282 5283
  //
  // Do not let reads from the cloned object float above the arraycopy.
5284 5285 5286 5287 5288 5289 5290 5291 5292 5293
  if (alloc != NULL) {
    // Do not let stores that initialize this object be reordered with
    // a subsequent store that would make this object accessible by
    // other threads.
    // Record what AllocateNode this StoreStore protects so that
    // escape analysis can go from the MemBarStoreStoreNode to the
    // AllocateNode and eliminate the MemBarStoreStoreNode if possible
    // based on the escape status of the AllocateNode.
    insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
  } else if (InsertMemBarAfterArraycopy)
D
duke 已提交
5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353
    insert_mem_bar(Op_MemBarCPUOrder);
}


// Helper function which determines if an arraycopy immediately follows
// an allocation, with no intervening tests or other escapes for the object.
AllocateArrayNode*
LibraryCallKit::tightly_coupled_allocation(Node* ptr,
                                           RegionNode* slow_region) {
  if (stopped())             return NULL;  // no fast path
  if (C->AliasLevel() == 0)  return NULL;  // no MergeMems around

  AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
  if (alloc == NULL)  return NULL;

  Node* rawmem = memory(Compile::AliasIdxRaw);
  // Is the allocation's memory state untouched?
  if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
    // Bail out if there have been raw-memory effects since the allocation.
    // (Example:  There might have been a call or safepoint.)
    return NULL;
  }
  rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
  if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
    return NULL;
  }

  // There must be no unexpected observers of this allocation.
  for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
    Node* obs = ptr->fast_out(i);
    if (obs != this->map()) {
      return NULL;
    }
  }

  // This arraycopy must unconditionally follow the allocation of the ptr.
  Node* alloc_ctl = ptr->in(0);
  assert(just_allocated_object(alloc_ctl) == ptr, "most recent allo");

  Node* ctl = control();
  while (ctl != alloc_ctl) {
    // There may be guards which feed into the slow_region.
    // Any other control flow means that we might not get a chance
    // to finish initializing the allocated object.
    if ((ctl->is_IfFalse() || ctl->is_IfTrue()) && ctl->in(0)->is_If()) {
      IfNode* iff = ctl->in(0)->as_If();
      Node* not_ctl = iff->proj_out(1 - ctl->as_Proj()->_con);
      assert(not_ctl != NULL && not_ctl != ctl, "found alternate");
      if (slow_region != NULL && slow_region->find_edge(not_ctl) >= 1) {
        ctl = iff->in(0);       // This test feeds the known slow_region.
        continue;
      }
      // One more try:  Various low-level checks bottom out in
      // uncommon traps.  If the debug-info of the trap omits
      // any reference to the allocation, as we've already
      // observed, then there can be no objection to the trap.
      bool found_trap = false;
      for (DUIterator_Fast jmax, j = not_ctl->fast_outs(jmax); j < jmax; j++) {
        Node* obs = not_ctl->fast_out(j);
        if (obs->in(0) == not_ctl && obs->is_Call() &&
T
twisti 已提交
5354
            (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
D
duke 已提交
5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412
          found_trap = true; break;
        }
      }
      if (found_trap) {
        ctl = iff->in(0);       // This test feeds a harmless uncommon trap.
        continue;
      }
    }
    return NULL;
  }

  // If we get this far, we have an allocation which immediately
  // precedes the arraycopy, and we can take over zeroing the new object.
  // The arraycopy will finish the initialization, and provide
  // a new control state to which we will anchor the destination pointer.

  return alloc;
}

// Helper for initialization of arrays, creating a ClearArray.
// It writes zero bits in [start..end), within the body of an array object.
// The memory effects are all chained onto the 'adr_type' alias category.
//
// Since the object is otherwise uninitialized, we are free
// to put a little "slop" around the edges of the cleared area,
// as long as it does not go back into the array's header,
// or beyond the array end within the heap.
//
// The lower edge can be rounded down to the nearest jint and the
// upper edge can be rounded up to the nearest MinObjAlignmentInBytes.
//
// Arguments:
//   adr_type           memory slice where writes are generated
//   dest               oop of the destination array
//   basic_elem_type    element type of the destination
//   slice_idx          array index of first element to store
//   slice_len          number of elements to store (or NULL)
//   dest_size          total size in bytes of the array object
//
// Exactly one of slice_len or dest_size must be non-NULL.
// If dest_size is non-NULL, zeroing extends to the end of the object.
// If slice_len is non-NULL, the slice_idx value must be a constant.
void
LibraryCallKit::generate_clear_array(const TypePtr* adr_type,
                                     Node* dest,
                                     BasicType basic_elem_type,
                                     Node* slice_idx,
                                     Node* slice_len,
                                     Node* dest_size) {
  // one or the other but not both of slice_len and dest_size:
  assert((slice_len != NULL? 1: 0) + (dest_size != NULL? 1: 0) == 1, "");
  if (slice_len == NULL)  slice_len = top();
  if (dest_size == NULL)  dest_size = top();

  // operate on this memory slice:
  Node* mem = memory(adr_type); // memory slice to operate on

  // scaling and rounding of indexes:
5413
  int scale = exact_log2(type2aelembytes(basic_elem_type));
D
duke 已提交
5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449
  int abase = arrayOopDesc::base_offset_in_bytes(basic_elem_type);
  int clear_low = (-1 << scale) & (BytesPerInt  - 1);
  int bump_bit  = (-1 << scale) & BytesPerInt;

  // determine constant starts and ends
  const intptr_t BIG_NEG = -128;
  assert(BIG_NEG + 2*abase < 0, "neg enough");
  intptr_t slice_idx_con = (intptr_t) find_int_con(slice_idx, BIG_NEG);
  intptr_t slice_len_con = (intptr_t) find_int_con(slice_len, BIG_NEG);
  if (slice_len_con == 0) {
    return;                     // nothing to do here
  }
  intptr_t start_con = (abase + (slice_idx_con << scale)) & ~clear_low;
  intptr_t end_con   = find_intptr_t_con(dest_size, -1);
  if (slice_idx_con >= 0 && slice_len_con >= 0) {
    assert(end_con < 0, "not two cons");
    end_con = round_to(abase + ((slice_idx_con + slice_len_con) << scale),
                       BytesPerLong);
  }

  if (start_con >= 0 && end_con >= 0) {
    // Constant start and end.  Simple.
    mem = ClearArrayNode::clear_memory(control(), mem, dest,
                                       start_con, end_con, &_gvn);
  } else if (start_con >= 0 && dest_size != top()) {
    // Constant start, pre-rounded end after the tail of the array.
    Node* end = dest_size;
    mem = ClearArrayNode::clear_memory(control(), mem, dest,
                                       start_con, end, &_gvn);
  } else if (start_con >= 0 && slice_len != top()) {
    // Constant start, non-constant end.  End needs rounding up.
    // End offset = round_up(abase + ((slice_idx_con + slice_len) << scale), 8)
    intptr_t end_base  = abase + (slice_idx_con << scale);
    int      end_round = (-1 << scale) & (BytesPerLong  - 1);
    Node*    end       = ConvI2X(slice_len);
    if (scale != 0)
5450
      end = _gvn.transform(new(C) LShiftXNode(end, intcon(scale) ));
D
duke 已提交
5451
    end_base += end_round;
5452 5453
    end = _gvn.transform(new(C) AddXNode(end, MakeConX(end_base)));
    end = _gvn.transform(new(C) AndXNode(end, MakeConX(~end_round)));
D
duke 已提交
5454 5455 5456 5457 5458 5459 5460 5461
    mem = ClearArrayNode::clear_memory(control(), mem, dest,
                                       start_con, end, &_gvn);
  } else if (start_con < 0 && dest_size != top()) {
    // Non-constant start, pre-rounded end after the tail of the array.
    // This is almost certainly a "round-to-end" operation.
    Node* start = slice_idx;
    start = ConvI2X(start);
    if (scale != 0)
5462 5463
      start = _gvn.transform(new(C) LShiftXNode( start, intcon(scale) ));
    start = _gvn.transform(new(C) AddXNode(start, MakeConX(abase)));
D
duke 已提交
5464 5465 5466 5467
    if ((bump_bit | clear_low) != 0) {
      int to_clear = (bump_bit | clear_low);
      // Align up mod 8, then store a jint zero unconditionally
      // just before the mod-8 boundary.
5468 5469 5470 5471 5472 5473
      if (((abase + bump_bit) & ~to_clear) - bump_bit
          < arrayOopDesc::length_offset_in_bytes() + BytesPerInt) {
        bump_bit = 0;
        assert((abase & to_clear) == 0, "array base must be long-aligned");
      } else {
        // Bump 'start' up to (or past) the next jint boundary:
5474
        start = _gvn.transform(new(C) AddXNode(start, MakeConX(bump_bit)));
5475 5476
        assert((abase & clear_low) == 0, "array base must be int-aligned");
      }
D
duke 已提交
5477
      // Round bumped 'start' down to jlong boundary in body of array.
5478
      start = _gvn.transform(new(C) AndXNode(start, MakeConX(~to_clear)));
5479 5480
      if (bump_bit != 0) {
        // Store a zero to the immediately preceding jint:
5481
        Node* x1 = _gvn.transform(new(C) AddXNode(start, MakeConX(-bump_bit)));
5482 5483 5484 5485
        Node* p1 = basic_plus_adr(dest, x1);
        mem = StoreNode::make(_gvn, control(), mem, p1, adr_type, intcon(0), T_INT);
        mem = _gvn.transform(mem);
      }
D
duke 已提交
5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506
    }
    Node* end = dest_size; // pre-rounded
    mem = ClearArrayNode::clear_memory(control(), mem, dest,
                                       start, end, &_gvn);
  } else {
    // Non-constant start, unrounded non-constant end.
    // (Nobody zeroes a random midsection of an array using this routine.)
    ShouldNotReachHere();       // fix caller
  }

  // Done.
  set_memory(mem, adr_type);
}


bool
LibraryCallKit::generate_block_arraycopy(const TypePtr* adr_type,
                                         BasicType basic_elem_type,
                                         AllocateNode* alloc,
                                         Node* src,  Node* src_offset,
                                         Node* dest, Node* dest_offset,
5507
                                         Node* dest_size, bool dest_uninitialized) {
D
duke 已提交
5508
  // See if there is an advantage from block transfer.
5509
  int scale = exact_log2(type2aelembytes(basic_elem_type));
D
duke 已提交
5510 5511 5512 5513 5514 5515
  if (scale >= LogBytesPerLong)
    return false;               // it is already a block transfer

  // Look at the alignment of the starting offsets.
  int abase = arrayOopDesc::base_offset_in_bytes(basic_elem_type);

5516 5517 5518
  intptr_t src_off_con  = (intptr_t) find_int_con(src_offset, -1);
  intptr_t dest_off_con = (intptr_t) find_int_con(dest_offset, -1);
  if (src_off_con < 0 || dest_off_con < 0)
D
duke 已提交
5519 5520 5521
    // At present, we can only understand constants.
    return false;

5522 5523 5524
  intptr_t src_off  = abase + (src_off_con  << scale);
  intptr_t dest_off = abase + (dest_off_con << scale);

D
duke 已提交
5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547
  if (((src_off | dest_off) & (BytesPerLong-1)) != 0) {
    // Non-aligned; too bad.
    // One more chance:  Pick off an initial 32-bit word.
    // This is a common case, since abase can be odd mod 8.
    if (((src_off | dest_off) & (BytesPerLong-1)) == BytesPerInt &&
        ((src_off ^ dest_off) & (BytesPerLong-1)) == 0) {
      Node* sptr = basic_plus_adr(src,  src_off);
      Node* dptr = basic_plus_adr(dest, dest_off);
      Node* sval = make_load(control(), sptr, TypeInt::INT, T_INT, adr_type);
      store_to_memory(control(), dptr, sval, T_INT, adr_type);
      src_off += BytesPerInt;
      dest_off += BytesPerInt;
    } else {
      return false;
    }
  }
  assert(src_off % BytesPerLong == 0, "");
  assert(dest_off % BytesPerLong == 0, "");

  // Do this copy by giant steps.
  Node* sptr  = basic_plus_adr(src,  src_off);
  Node* dptr  = basic_plus_adr(dest, dest_off);
  Node* countx = dest_size;
5548 5549
  countx = _gvn.transform(new (C) SubXNode(countx, MakeConX(dest_off)));
  countx = _gvn.transform(new (C) URShiftXNode(countx, intcon(LogBytesPerLong)));
D
duke 已提交
5550 5551 5552

  bool disjoint_bases = true;   // since alloc != NULL
  generate_unchecked_arraycopy(adr_type, T_LONG, disjoint_bases,
5553
                               sptr, NULL, dptr, NULL, countx, dest_uninitialized);
D
duke 已提交
5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565

  return true;
}


// Helper function; generates code for the slow case.
// We make a call to a runtime method which emulates the native method,
// but without the native wrapper overhead.
void
LibraryCallKit::generate_slow_arraycopy(const TypePtr* adr_type,
                                        Node* src,  Node* src_offset,
                                        Node* dest, Node* dest_offset,
5566 5567
                                        Node* copy_length, bool dest_uninitialized) {
  assert(!dest_uninitialized, "Invariant");
D
duke 已提交
5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584
  Node* call = make_runtime_call(RC_NO_LEAF | RC_UNCOMMON,
                                 OptoRuntime::slow_arraycopy_Type(),
                                 OptoRuntime::slow_arraycopy_Java(),
                                 "slow_arraycopy", adr_type,
                                 src, src_offset, dest, dest_offset,
                                 copy_length);

  // Handle exceptions thrown by this fellow:
  make_slow_call_ex(call, env()->Throwable_klass(), false);
}

// Helper function; generates code for cases requiring runtime checks.
Node*
LibraryCallKit::generate_checkcast_arraycopy(const TypePtr* adr_type,
                                             Node* dest_elem_klass,
                                             Node* src,  Node* src_offset,
                                             Node* dest, Node* dest_offset,
5585
                                             Node* copy_length, bool dest_uninitialized) {
D
duke 已提交
5586 5587
  if (stopped())  return NULL;

5588
  address copyfunc_addr = StubRoutines::checkcast_arraycopy(dest_uninitialized);
D
duke 已提交
5589 5590 5591 5592 5593 5594 5595 5596
  if (copyfunc_addr == NULL) { // Stub was not generated, go slow path.
    return NULL;
  }

  // Pick out the parameters required to perform a store-check
  // for the target array.  This is an optimistic check.  It will
  // look in each non-null element's class, at the desired klass's
  // super_check_offset, for the desired klass.
5597
  int sco_offset = in_bytes(Klass::super_check_offset_offset());
D
duke 已提交
5598
  Node* p3 = basic_plus_adr(dest_elem_klass, sco_offset);
5599
  Node* n3 = new(C) LoadINode(NULL, memory(p3), p3, _gvn.type(p3)->is_ptr());
5600
  Node* check_offset = ConvI2X(_gvn.transform(n3));
D
duke 已提交
5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616
  Node* check_value  = dest_elem_klass;

  Node* src_start  = array_element_address(src,  src_offset,  T_OBJECT);
  Node* dest_start = array_element_address(dest, dest_offset, T_OBJECT);

  // (We know the arrays are never conjoint, because their types differ.)
  Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
                                 OptoRuntime::checkcast_arraycopy_Type(),
                                 copyfunc_addr, "checkcast_arraycopy", adr_type,
                                 // five arguments, of which two are
                                 // intptr_t (jlong in LP64)
                                 src_start, dest_start,
                                 copy_length XTOP,
                                 check_offset XTOP,
                                 check_value);

5617
  return _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
D
duke 已提交
5618 5619 5620 5621 5622 5623 5624 5625
}


// Helper function; generates code for cases requiring runtime checks.
Node*
LibraryCallKit::generate_generic_arraycopy(const TypePtr* adr_type,
                                           Node* src,  Node* src_offset,
                                           Node* dest, Node* dest_offset,
5626 5627
                                           Node* copy_length, bool dest_uninitialized) {
  assert(!dest_uninitialized, "Invariant");
D
duke 已提交
5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638
  if (stopped())  return NULL;
  address copyfunc_addr = StubRoutines::generic_arraycopy();
  if (copyfunc_addr == NULL) { // Stub was not generated, go slow path.
    return NULL;
  }

  Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
                    OptoRuntime::generic_arraycopy_Type(),
                    copyfunc_addr, "generic_arraycopy", adr_type,
                    src, src_offset, dest, dest_offset, copy_length);

5639
  return _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
D
duke 已提交
5640 5641 5642 5643 5644 5645 5646 5647 5648
}

// Helper function; generates the fast out-of-line call to an arraycopy stub.
void
LibraryCallKit::generate_unchecked_arraycopy(const TypePtr* adr_type,
                                             BasicType basic_elem_type,
                                             bool disjoint_bases,
                                             Node* src,  Node* src_offset,
                                             Node* dest, Node* dest_offset,
5649
                                             Node* copy_length, bool dest_uninitialized) {
D
duke 已提交
5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663
  if (stopped())  return;               // nothing to do

  Node* src_start  = src;
  Node* dest_start = dest;
  if (src_offset != NULL || dest_offset != NULL) {
    assert(src_offset != NULL && dest_offset != NULL, "");
    src_start  = array_element_address(src,  src_offset,  basic_elem_type);
    dest_start = array_element_address(dest, dest_offset, basic_elem_type);
  }

  // Figure out which arraycopy runtime method to call.
  const char* copyfunc_name = "arraycopy";
  address     copyfunc_addr =
      basictype2arraycopy(basic_elem_type, src_offset, dest_offset,
5664
                          disjoint_bases, copyfunc_name, dest_uninitialized);
D
duke 已提交
5665 5666 5667 5668 5669 5670 5671

  // Call it.  Note that the count_ix value is not scaled to a byte-size.
  make_runtime_call(RC_LEAF|RC_NO_FP,
                    OptoRuntime::fast_arraycopy_Type(),
                    copyfunc_addr, copyfunc_name, adr_type,
                    src_start, dest_start, copy_length XTOP);
}
5672

5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713
//-------------inline_encodeISOArray-----------------------------------
// encode char[] to byte[] in ISO_8859_1
bool LibraryCallKit::inline_encodeISOArray() {
  assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
  // no receiver since it is static method
  Node *src         = argument(0);
  Node *src_offset  = argument(1);
  Node *dst         = argument(2);
  Node *dst_offset  = argument(3);
  Node *length      = argument(4);

  const Type* src_type = src->Value(&_gvn);
  const Type* dst_type = dst->Value(&_gvn);
  const TypeAryPtr* top_src = src_type->isa_aryptr();
  const TypeAryPtr* top_dest = dst_type->isa_aryptr();
  if (top_src  == NULL || top_src->klass()  == NULL ||
      top_dest == NULL || top_dest->klass() == NULL) {
    // failed array check
    return false;
  }

  // Figure out the size and type of the elements we will be copying.
  BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
  BasicType dst_elem = dst_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
  if (src_elem != T_CHAR || dst_elem != T_BYTE) {
    return false;
  }
  Node* src_start = array_element_address(src, src_offset, src_elem);
  Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
  // 'src_start' points to src array + scaled offset
  // 'dst_start' points to dst array + scaled offset

  const TypeAryPtr* mtype = TypeAryPtr::BYTES;
  Node* enc = new (C) EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length);
  enc = _gvn.transform(enc);
  Node* res_mem = _gvn.transform(new (C) SCMemProjNode(enc));
  set_memory(res_mem, mtype);
  set_result(enc);
  return true;
}

5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824
/**
 * Calculate CRC32 for byte.
 * int java.util.zip.CRC32.update(int crc, int b)
 */
bool LibraryCallKit::inline_updateCRC32() {
  assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
  assert(callee()->signature()->size() == 2, "update has 2 parameters");
  // no receiver since it is static method
  Node* crc  = argument(0); // type: int
  Node* b    = argument(1); // type: int

  /*
   *    int c = ~ crc;
   *    b = timesXtoThe32[(b ^ c) & 0xFF];
   *    b = b ^ (c >>> 8);
   *    crc = ~b;
   */

  Node* M1 = intcon(-1);
  crc = _gvn.transform(new (C) XorINode(crc, M1));
  Node* result = _gvn.transform(new (C) XorINode(crc, b));
  result = _gvn.transform(new (C) AndINode(result, intcon(0xFF)));

  Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
  Node* offset = _gvn.transform(new (C) LShiftINode(result, intcon(0x2)));
  Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
  result = make_load(control(), adr, TypeInt::INT, T_INT);

  crc = _gvn.transform(new (C) URShiftINode(crc, intcon(8)));
  result = _gvn.transform(new (C) XorINode(crc, result));
  result = _gvn.transform(new (C) XorINode(result, M1));
  set_result(result);
  return true;
}

/**
 * Calculate CRC32 for byte[] array.
 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
 */
bool LibraryCallKit::inline_updateBytesCRC32() {
  assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
  assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
  // no receiver since it is static method
  Node* crc     = argument(0); // type: int
  Node* src     = argument(1); // type: oop
  Node* offset  = argument(2); // type: int
  Node* length  = argument(3); // type: int

  const Type* src_type = src->Value(&_gvn);
  const TypeAryPtr* top_src = src_type->isa_aryptr();
  if (top_src  == NULL || top_src->klass()  == NULL) {
    // failed array check
    return false;
  }

  // Figure out the size and type of the elements we will be copying.
  BasicType src_elem = src_type->isa_aryptr()->klass()->as_array_klass()->element_type()->basic_type();
  if (src_elem != T_BYTE) {
    return false;
  }

  // 'src_start' points to src array + scaled offset
  Node* src_start = array_element_address(src, offset, src_elem);

  // We assume that range check is done by caller.
  // TODO: generate range check (offset+length < src.length) in debug VM.

  // Call the stub.
  address stubAddr = StubRoutines::updateBytesCRC32();
  const char *stubName = "updateBytesCRC32";

  Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
                                 stubAddr, stubName, TypePtr::BOTTOM,
                                 crc, src_start, length);
  Node* result = _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
  set_result(result);
  return true;
}

/**
 * Calculate CRC32 for ByteBuffer.
 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
 */
bool LibraryCallKit::inline_updateByteBufferCRC32() {
  assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
  assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
  // no receiver since it is static method
  Node* crc     = argument(0); // type: int
  Node* src     = argument(1); // type: long
  Node* offset  = argument(3); // type: int
  Node* length  = argument(4); // type: int

  src = ConvL2X(src);  // adjust Java long to machine word
  Node* base = _gvn.transform(new (C) CastX2PNode(src));
  offset = ConvI2X(offset);

  // 'src_start' points to src array + scaled offset
  Node* src_start = basic_plus_adr(top(), base, offset);

  // Call the stub.
  address stubAddr = StubRoutines::updateBytesCRC32();
  const char *stubName = "updateBytesCRC32";

  Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
                                 stubAddr, stubName, TypePtr::BOTTOM,
                                 crc, src_start, length);
  Node* result = _gvn.transform(new (C) ProjNode(call, TypeFunc::Parms));
  set_result(result);
  return true;
}

5825
//----------------------------inline_reference_get----------------------------
5826
// public T java.lang.ref.Reference.get();
5827
bool LibraryCallKit::inline_reference_get() {
5828 5829
  const int referent_offset = java_lang_ref_Reference::referent_offset;
  guarantee(referent_offset > 0, "should have already been set");
5830

5831 5832
  // Get the argument:
  Node* reference_obj = null_check_receiver();
5833 5834
  if (stopped()) return true;

5835
  Node* adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);
5836 5837 5838 5839 5840

  ciInstanceKlass* klass = env()->Object_klass();
  const TypeOopPtr* object_type = TypeOopPtr::make_from_klass(klass);

  Node* no_ctrl = NULL;
5841
  Node* result = make_load(no_ctrl, adr, object_type, T_OBJECT);
5842 5843 5844 5845

  // Use the pre-barrier to record the value in the referent field
  pre_barrier(false /* do_load */,
              control(),
5846
              NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5847 5848 5849
              result /* pre_val */,
              T_OBJECT);

5850 5851 5852 5853
  // Add memory barrier to prevent commoning reads from this field
  // across safepoint since GC can change its value.
  insert_mem_bar(Op_MemBarCPUOrder);

5854
  set_result(result);
5855 5856
  return true;
}
5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882 5883 5884 5885 5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910


Node * LibraryCallKit::load_field_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
                                              bool is_exact=true, bool is_static=false) {

  const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
  assert(tinst != NULL, "obj is null");
  assert(tinst->klass()->is_loaded(), "obj is not loaded");
  assert(!is_exact || tinst->klass_is_exact(), "klass not exact");

  ciField* field = tinst->klass()->as_instance_klass()->get_field_by_name(ciSymbol::make(fieldName),
                                                                          ciSymbol::make(fieldTypeString),
                                                                          is_static);
  if (field == NULL) return (Node *) NULL;
  assert (field != NULL, "undefined field");

  // Next code  copied from Parse::do_get_xxx():

  // Compute address and memory type.
  int offset  = field->offset_in_bytes();
  bool is_vol = field->is_volatile();
  ciType* field_klass = field->type();
  assert(field_klass->is_loaded(), "should be loaded");
  const TypePtr* adr_type = C->alias_type(field)->adr_type();
  Node *adr = basic_plus_adr(fromObj, fromObj, offset);
  BasicType bt = field->layout_type();

  // Build the resultant type of the load
  const Type *type = TypeOopPtr::make_from_klass(field_klass->as_klass());

  // Build the load.
  Node* loadedField = make_load(NULL, adr, type, bt, adr_type, is_vol);
  return loadedField;
}


//------------------------------inline_aescrypt_Block-----------------------
bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
  address stubAddr;
  const char *stubName;
  assert(UseAES, "need AES instruction support");

  switch(id) {
  case vmIntrinsics::_aescrypt_encryptBlock:
    stubAddr = StubRoutines::aescrypt_encryptBlock();
    stubName = "aescrypt_encryptBlock";
    break;
  case vmIntrinsics::_aescrypt_decryptBlock:
    stubAddr = StubRoutines::aescrypt_decryptBlock();
    stubName = "aescrypt_decryptBlock";
    break;
  }
  if (stubAddr == NULL) return false;

5911 5912 5913 5914 5915
  Node* aescrypt_object = argument(0);
  Node* src             = argument(1);
  Node* src_offset      = argument(2);
  Node* dest            = argument(3);
  Node* dest_offset     = argument(4);
5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965

  // (1) src and dest are arrays.
  const Type* src_type = src->Value(&_gvn);
  const Type* dest_type = dest->Value(&_gvn);
  const TypeAryPtr* top_src = src_type->isa_aryptr();
  const TypeAryPtr* top_dest = dest_type->isa_aryptr();
  assert (top_src  != NULL && top_src->klass()  != NULL &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");

  // for the quick and dirty code we will skip all the checks.
  // we are just trying to get the call to be generated.
  Node* src_start  = src;
  Node* dest_start = dest;
  if (src_offset != NULL || dest_offset != NULL) {
    assert(src_offset != NULL && dest_offset != NULL, "");
    src_start  = array_element_address(src,  src_offset,  T_BYTE);
    dest_start = array_element_address(dest, dest_offset, T_BYTE);
  }

  // now need to get the start of its expanded key array
  // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
  Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
  if (k_start == NULL) return false;

  // Call the stub.
  make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
                    stubAddr, stubName, TypePtr::BOTTOM,
                    src_start, dest_start, k_start);

  return true;
}

//------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
  address stubAddr;
  const char *stubName;

  assert(UseAES, "need AES instruction support");

  switch(id) {
  case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
    stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
    stubName = "cipherBlockChaining_encryptAESCrypt";
    break;
  case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
    stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
    stubName = "cipherBlockChaining_decryptAESCrypt";
    break;
  }
  if (stubAddr == NULL) return false;

5966 5967 5968 5969 5970 5971
  Node* cipherBlockChaining_object = argument(0);
  Node* src                        = argument(1);
  Node* src_offset                 = argument(2);
  Node* len                        = argument(3);
  Node* dest                       = argument(4);
  Node* dest_offset                = argument(5);
5972 5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044 6045 6046 6047 6048 6049 6050 6051 6052 6053

  // (1) src and dest are arrays.
  const Type* src_type = src->Value(&_gvn);
  const Type* dest_type = dest->Value(&_gvn);
  const TypeAryPtr* top_src = src_type->isa_aryptr();
  const TypeAryPtr* top_dest = dest_type->isa_aryptr();
  assert (top_src  != NULL && top_src->klass()  != NULL
          &&  top_dest != NULL && top_dest->klass() != NULL, "args are strange");

  // checks are the responsibility of the caller
  Node* src_start  = src;
  Node* dest_start = dest;
  if (src_offset != NULL || dest_offset != NULL) {
    assert(src_offset != NULL && dest_offset != NULL, "");
    src_start  = array_element_address(src,  src_offset,  T_BYTE);
    dest_start = array_element_address(dest, dest_offset, T_BYTE);
  }

  // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
  // (because of the predicated logic executed earlier).
  // so we cast it here safely.
  // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java

  Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);
  if (embeddedCipherObj == NULL) return false;

  // cast it to what we know it will be at runtime
  const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
  assert(tinst != NULL, "CBC obj is null");
  assert(tinst->klass()->is_loaded(), "CBC obj is not loaded");
  ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
  if (!klass_AESCrypt->is_loaded()) return false;

  ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
  const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
  const TypeOopPtr* xtype = aklass->as_instance_type();
  Node* aescrypt_object = new(C) CheckCastPPNode(control(), embeddedCipherObj, xtype);
  aescrypt_object = _gvn.transform(aescrypt_object);

  // we need to get the start of the aescrypt_object's expanded key array
  Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
  if (k_start == NULL) return false;

  // similarly, get the start address of the r vector
  Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B", /*is_exact*/ false);
  if (objRvec == NULL) return false;
  Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);

  // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
  make_runtime_call(RC_LEAF|RC_NO_FP,
                    OptoRuntime::cipherBlockChaining_aescrypt_Type(),
                    stubAddr, stubName, TypePtr::BOTTOM,
                    src_start, dest_start, k_start, r_start, len);

  // return is void so no result needs to be pushed

  return true;
}

//------------------------------get_key_start_from_aescrypt_object-----------------------
Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
  Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I", /*is_exact*/ false);
  assert (objAESCryptKey != NULL, "wrong version of com.sun.crypto.provider.AESCrypt");
  if (objAESCryptKey == NULL) return (Node *) NULL;

  // now have the array, need to get the start address of the K array
  Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
  return k_start;
}

//----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
// Return node representing slow path of predicate check.
// the pseudo code we want to emulate with this predicate is:
// for encryption:
//    if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
// for decryption:
//    if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
//    note cipher==plain is more conservative than the original java code but that's OK
//
Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
  // First, check receiver for NULL since it is virtual method.
  Node* objCBC = argument(0);
6054
  objCBC = null_check(objCBC);
6055 6056 6057 6058 6059 6060 6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090 6091 6092 6093

  if (stopped()) return NULL; // Always NULL

  // Load embeddedCipher field of CipherBlockChaining object.
  Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;", /*is_exact*/ false);

  // get AESCrypt klass for instanceOf check
  // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
  // will have same classloader as CipherBlockChaining object
  const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
  assert(tinst != NULL, "CBCobj is null");
  assert(tinst->klass()->is_loaded(), "CBCobj is not loaded");

  // we want to do an instanceof comparison against the AESCrypt class
  ciKlass* klass_AESCrypt = tinst->klass()->as_instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
  if (!klass_AESCrypt->is_loaded()) {
    // if AESCrypt is not even loaded, we never take the intrinsic fast path
    Node* ctrl = control();
    set_control(top()); // no regular fast path
    return ctrl;
  }
  ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();

  Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
  Node* cmp_instof  = _gvn.transform(new (C) CmpINode(instof, intcon(1)));
  Node* bool_instof  = _gvn.transform(new (C) BoolNode(cmp_instof, BoolTest::ne));

  Node* instof_false = generate_guard(bool_instof, NULL, PROB_MIN);

  // for encryption, we are done
  if (!decrypting)
    return instof_false;  // even if it is NULL

  // for decryption, we need to add a further check to avoid
  // taking the intrinsic path when cipher and plain are the same
  // see the original java code for why.
  RegionNode* region = new(C) RegionNode(3);
  region->init_req(1, instof_false);
  Node* src = argument(1);
6094
  Node* dest = argument(4);
6095 6096 6097 6098 6099 6100 6101 6102
  Node* cmp_src_dest = _gvn.transform(new (C) CmpPNode(src, dest));
  Node* bool_src_dest = _gvn.transform(new (C) BoolNode(cmp_src_dest, BoolTest::eq));
  Node* src_dest_conjoint = generate_guard(bool_src_dest, NULL, PROB_MIN);
  region->init_req(2, src_dest_conjoint);

  record_for_igvn(region);
  return _gvn.transform(region);
}