library_call.cpp 215.1 KB
Newer Older
D
duke 已提交
1
/*
2
 * Copyright (c) 1999, 2011, 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 35 36 37 38 39 40
#include "compiler/compileLog.hpp"
#include "oops/objArrayKlass.hpp"
#include "opto/addnode.hpp"
#include "opto/callGenerator.hpp"
#include "opto/cfgnode.hpp"
#include "opto/idealKit.hpp"
#include "opto/mulnode.hpp"
#include "opto/parse.hpp"
#include "opto/runtime.hpp"
#include "opto/subnode.hpp"
#include "prims/nativeLookup.hpp"
#include "runtime/sharedRuntime.hpp"
D
duke 已提交
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100

class LibraryIntrinsic : public InlineCallGenerator {
  // Extend the set of intrinsics known to the runtime:
 public:
 private:
  bool             _is_virtual;
  vmIntrinsics::ID _intrinsic_id;

 public:
  LibraryIntrinsic(ciMethod* m, bool is_virtual, vmIntrinsics::ID id)
    : InlineCallGenerator(m),
      _is_virtual(is_virtual),
      _intrinsic_id(id)
  {
  }
  virtual bool is_intrinsic() const { return true; }
  virtual bool is_virtual()   const { return _is_virtual; }
  virtual JVMState* generate(JVMState* jvms);
  vmIntrinsics::ID intrinsic_id() const { return _intrinsic_id; }
};


// Local helper class for LibraryIntrinsic:
class LibraryCallKit : public GraphKit {
 private:
  LibraryIntrinsic* _intrinsic;   // the library intrinsic being called

 public:
  LibraryCallKit(JVMState* caller, LibraryIntrinsic* intrinsic)
    : GraphKit(caller),
      _intrinsic(intrinsic)
  {
  }

  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(); }
  ciSignature*      signature() const    { return callee()->signature(); }
  int               arg_size()  const    { return callee()->arg_size(); }

  bool try_to_inline();

  // Helper functions to inline natives
  void push_result(RegionNode* region, PhiNode* value);
  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,
101
                              bool disjoint_bases, const char* &name, bool dest_uninitialized);
D
duke 已提交
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149
  Node* load_mirror_from_klass(Node* klass);
  Node* load_klass_from_mirror_common(Node* mirror, bool never_see_null,
                                      int nargs,
                                      RegionNode* region, int null_path,
                                      int offset);
  Node* load_klass_from_mirror(Node* mirror, bool never_see_null, int nargs,
                               RegionNode* region, int null_path) {
    int offset = java_lang_Class::klass_offset_in_bytes();
    return load_klass_from_mirror_common(mirror, never_see_null, nargs,
                                         region, null_path,
                                         offset);
  }
  Node* load_array_klass_from_mirror(Node* mirror, bool never_see_null,
                                     int nargs,
                                     RegionNode* region, int null_path) {
    int offset = java_lang_Class::array_klass_offset_in_bytes();
    return load_klass_from_mirror_common(mirror, never_see_null, nargs,
                                         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);
  }

150
  Node* make_string_method_node(int opcode, Node* str1, Node* cnt1, Node* str2, Node* cnt2);
D
duke 已提交
151 152 153
  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
154
  bool inline_string_equals();
D
duke 已提交
155 156 157 158 159 160 161 162 163 164 165 166 167 168
  Node* pop_math_arg();
  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);
  bool inline_trans(vmIntrinsics::ID id);
  bool inline_abs(vmIntrinsics::ID id);
  bool inline_sqrt(vmIntrinsics::ID id);
  bool inline_pow(vmIntrinsics::ID id);
  bool inline_exp(vmIntrinsics::ID id);
  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);
169 170 171 172
  // Helper for inline_unsafe_access.
  // Generates the guards that check whether the result of
  // Unsafe.getObject should be recorded in an SATB log buffer.
  void insert_g1_pre_barrier(Node* base_oop, Node* offset, Node* pre_val);
D
duke 已提交
173 174 175 176 177 178 179 180 181 182 183 184 185
  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);
  bool inline_unsafe_allocate();
  bool inline_unsafe_copyMemory();
  bool inline_native_currentThread();
  bool inline_native_time_funcs(bool isNano);
  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);
186
  bool inline_array_equals();
187
  void copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array, bool card_mark);
D
duke 已提交
188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
  bool inline_native_clone(bool is_virtual);
  bool inline_native_Reflection_getCallerClass();
  bool inline_native_AtomicLong_get();
  bool inline_native_AtomicLong_attemptUpdate();
  bool is_method_invoke_or_aux_frame(JVMState* jvms);
  // 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,
220
                                Node* dest_size, bool dest_uninitialized);
D
duke 已提交
221 222 223
  void generate_slow_arraycopy(const TypePtr* adr_type,
                               Node* src,  Node* src_offset,
                               Node* dest, Node* dest_offset,
224
                               Node* copy_length, bool dest_uninitialized);
D
duke 已提交
225 226 227 228
  Node* generate_checkcast_arraycopy(const TypePtr* adr_type,
                                     Node* dest_elem_klass,
                                     Node* src,  Node* src_offset,
                                     Node* dest, Node* dest_offset,
229
                                     Node* copy_length, bool dest_uninitialized);
D
duke 已提交
230 231 232
  Node* generate_generic_arraycopy(const TypePtr* adr_type,
                                   Node* src,  Node* src_offset,
                                   Node* dest, Node* dest_offset,
233
                                   Node* copy_length, bool dest_uninitialized);
D
duke 已提交
234 235 236 237 238
  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,
239
                                    Node* copy_length, bool dest_uninitialized);
D
duke 已提交
240 241 242
  bool inline_unsafe_CAS(BasicType type);
  bool inline_unsafe_ordered_store(BasicType type);
  bool inline_fp_conversions(vmIntrinsics::ID id);
243 244
  bool inline_numberOfLeadingZeros(vmIntrinsics::ID id);
  bool inline_numberOfTrailingZeros(vmIntrinsics::ID id);
245
  bool inline_bitCount(vmIntrinsics::ID id);
D
duke 已提交
246
  bool inline_reverseBytes(vmIntrinsics::ID id);
247 248

  bool inline_reference_get();
D
duke 已提交
249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
};


//---------------------------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
287
    case vmIntrinsics::_equals:
288
    case vmIntrinsics::_equalsC:
D
duke 已提交
289 290 291 292 293 294 295 296 297 298 299 300 301
      break;  // InlineNatives does not control String.compareTo
    default:
      return NULL;
    }
  }

  switch (id) {
  case vmIntrinsics::_compareTo:
    if (!SpecialStringCompareTo)  return NULL;
    break;
  case vmIntrinsics::_indexOf:
    if (!SpecialStringIndexOf)  return NULL;
    break;
C
cfang 已提交
302 303 304
  case vmIntrinsics::_equals:
    if (!SpecialStringEquals)  return NULL;
    break;
305 306 307
  case vmIntrinsics::_equalsC:
    if (!SpecialArraysEquals)  return NULL;
    break;
D
duke 已提交
308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339
  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;
    break;
  case vmIntrinsics::_clone:
  case vmIntrinsics::_copyOf:
  case vmIntrinsics::_copyOfRange:
    if (!InlineObjectCopy)  return NULL;
    // These also use the arraycopy intrinsic mechanism:
    if (!InlineArrayCopy)  return NULL;
    break;
  case vmIntrinsics::_checkIndex:
    // We do not intrinsify this.  The optimizer does fine with it.
    return NULL;

  case vmIntrinsics::_get_AtomicLong:
  case vmIntrinsics::_attemptUpdate:
    if (!InlineAtomicLong)  return NULL;
    break;

  case vmIntrinsics::_getCallerClass:
    if (!UseNewReflection)  return NULL;
    if (!InlineReflectionGetCallerClass)  return NULL;
    if (!JDK_Version::is_gte_jdk14x_version())  return NULL;
    break;

340 341 342 343 344
  case vmIntrinsics::_bitCount_i:
  case vmIntrinsics::_bitCount_l:
    if (!UsePopCountInstruction)  return NULL;
    break;

345 346 347 348 349 350 351 352
  case vmIntrinsics::_Reference_get:
    // It is only when G1 is enabled that we absolutely
    // need to use the intrinsic version of Reference.get()
    // so that the value in the referent field, if necessary,
    // can be registered by the pre-barrier code.
    if (!UseG1GC) return NULL;
    break;

D
duke 已提交
353
 default:
354 355
    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
356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403
    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;
  }

  return new LibraryIntrinsic(m, is_virtual, (vmIntrinsics::ID) id);
}

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

JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
  LibraryCallKit kit(jvms, this);
  Compile* C = kit.C;
  int nodes = C->unique();
#ifndef PRODUCT
  if ((PrintIntrinsics || PrintInlining NOT_PRODUCT( || PrintOptoInlining) ) && Verbose) {
    char buf[1000];
    const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
    tty->print_cr("Intrinsic %s", str);
  }
#endif
404

D
duke 已提交
405 406
  if (kit.try_to_inline()) {
    if (PrintIntrinsics || PrintInlining NOT_PRODUCT( || PrintOptoInlining) ) {
407
      CompileTask::print_inlining(kit.callee(), jvms->depth() - 1, kit.bci(), is_virtual() ? "(intrinsic, virtual)" : "(intrinsic)");
D
duke 已提交
408 409 410 411 412 413 414 415 416 417 418 419
    }
    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);
    }
    return kit.transfer_exceptions_into_jvms();
  }

  if (PrintIntrinsics) {
420 421 422 423 424 425 426 427 428 429
    if (jvms->has_method()) {
      // Not a root compile.
      tty->print("Did not inline intrinsic %s%s at bci:%d in",
                 vmIntrinsics::name_at(intrinsic_id()),
                 (is_virtual() ? " (virtual)" : ""), kit.bci());
      kit.caller()->print_short_name(tty);
      tty->print_cr(" (%d bytes)", kit.caller()->code_size());
    } else {
      // Root compile
      tty->print("Did not generate intrinsic %s%s at bci:%d in",
430 431
               vmIntrinsics::name_at(intrinsic_id()),
               (is_virtual() ? " (virtual)" : ""), kit.bci());
432
    }
D
duke 已提交
433 434 435 436 437 438 439 440 441 442 443
  }
  C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
  return NULL;
}

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;

444 445 446 447 448 449 450 451
  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(), "");

D
duke 已提交
452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482
  switch (intrinsic_id()) {
  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();

  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:
  case vmIntrinsics::_dpow:
    return inline_math_native(intrinsic_id());

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

  case vmIntrinsics::_arraycopy:
    return inline_arraycopy();

  case vmIntrinsics::_compareTo:
    return inline_string_compareTo();
  case vmIntrinsics::_indexOf:
    return inline_string_indexOf();
C
cfang 已提交
483 484
  case vmIntrinsics::_equals:
    return inline_string_equals();
D
duke 已提交
485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639

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

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

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

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

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

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

  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_CAS(T_OBJECT);
  case vmIntrinsics::_compareAndSwapInt:
    return inline_unsafe_CAS(T_INT);
  case vmIntrinsics::_compareAndSwapLong:
    return inline_unsafe_CAS(T_LONG);

  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::_currentThread:
    return inline_native_currentThread();
  case vmIntrinsics::_isInterrupted:
    return inline_native_isInterrupted();

  case vmIntrinsics::_currentTimeMillis:
    return inline_native_time_funcs(false);
  case vmIntrinsics::_nanoTime:
    return inline_native_time_funcs(true);
  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);
640 641
  case vmIntrinsics::_equalsC:
    return inline_array_equals();
D
duke 已提交
642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665
  case vmIntrinsics::_clone:
    return inline_native_clone(intrinsic()->is_virtual());

  case vmIntrinsics::_isAssignableFrom:
    return inline_native_subtype_check();

  case vmIntrinsics::_isInstance:
  case vmIntrinsics::_getModifiers:
  case vmIntrinsics::_isInterface:
  case vmIntrinsics::_isArray:
  case vmIntrinsics::_isPrimitive:
  case vmIntrinsics::_getSuperclass:
  case vmIntrinsics::_getComponentType:
  case vmIntrinsics::_getClassAccessFlags:
    return inline_native_Class_query(intrinsic_id());

  case vmIntrinsics::_floatToRawIntBits:
  case vmIntrinsics::_floatToIntBits:
  case vmIntrinsics::_intBitsToFloat:
  case vmIntrinsics::_doubleToRawLongBits:
  case vmIntrinsics::_doubleToLongBits:
  case vmIntrinsics::_longBitsToDouble:
    return inline_fp_conversions(intrinsic_id());

666 667 668 669 670 671 672 673
  case vmIntrinsics::_numberOfLeadingZeros_i:
  case vmIntrinsics::_numberOfLeadingZeros_l:
    return inline_numberOfLeadingZeros(intrinsic_id());

  case vmIntrinsics::_numberOfTrailingZeros_i:
  case vmIntrinsics::_numberOfTrailingZeros_l:
    return inline_numberOfTrailingZeros(intrinsic_id());

674 675 676 677
  case vmIntrinsics::_bitCount_i:
  case vmIntrinsics::_bitCount_l:
    return inline_bitCount(intrinsic_id());

D
duke 已提交
678 679
  case vmIntrinsics::_reverseBytes_i:
  case vmIntrinsics::_reverseBytes_l:
680 681
  case vmIntrinsics::_reverseBytes_s:
  case vmIntrinsics::_reverseBytes_c:
D
duke 已提交
682 683 684 685 686 687 688 689 690 691
    return inline_reverseBytes((vmIntrinsics::ID) intrinsic_id());

  case vmIntrinsics::_get_AtomicLong:
    return inline_native_AtomicLong_get();
  case vmIntrinsics::_attemptUpdate:
    return inline_native_AtomicLong_attemptUpdate();

  case vmIntrinsics::_getCallerClass:
    return inline_native_Reflection_getCallerClass();

692 693 694
  case vmIntrinsics::_Reference_get:
    return inline_reference_get();

D
duke 已提交
695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845
  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;
  }
}

//------------------------------push_result------------------------------
// Helper function for finishing intrinsics.
void LibraryCallKit::push_result(RegionNode* region, PhiNode* value) {
  record_for_igvn(region);
  set_control(_gvn.transform(region));
  BasicType value_type = value->type()->basic_type();
  push_node(value_type, _gvn.transform(value));
}

//------------------------------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);

  Node* if_slow = _gvn.transform( new (C, 1) IfTrueNode(iff) );
  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);

  Node* if_fast = _gvn.transform( new (C, 1) IfFalseNode(iff) );
  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
  Node* cmp_lt = _gvn.transform( new (C, 3) CmpINode(index, intcon(0)) );
  Node* bol_lt = _gvn.transform( new (C, 2) BoolNode(cmp_lt, BoolTest::lt) );
  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.
    Node* ccast = new (C, 2) CastIINode(index, TypeInt::POS);
    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
  Node* cmp_le = _gvn.transform( new (C, 3) CmpINode(index, intcon(0)) );
  BoolTest::mask le_or_eq = (never_negative ? BoolTest::eq : BoolTest::le);
  Node* bol_le = _gvn.transform( new (C, 2) BoolNode(cmp_le, le_or_eq) );
  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.
    Node* ccast = new (C, 2) CastIINode(index, TypeInt::POS1);
    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;
  if (zero_offset && _gvn.eqv_uncast(subseq_length, array_length))
    return NULL;                // common case of whole-array copy
  Node* last = subseq_length;
  if (!zero_offset)             // last += offset
    last = _gvn.transform( new (C, 3) AddINode(last, offset));
  Node* cmp_lt = _gvn.transform( new (C, 3) CmpUNode(array_length, last) );
  Node* bol_lt = _gvn.transform( new (C, 2) BoolNode(cmp_lt, BoolTest::lt) );
  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);
  Node* thread = _gvn.transform(new (C, 1) ThreadLocalNode());
  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;
}


846 847 848 849 850 851 852 853 854 855
//------------------------------make_string_method_node------------------------
// Helper method for String intrinsic finctions.
Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1, Node* cnt1, Node* str2, Node* cnt2) {
  const int value_offset  = java_lang_String::value_offset_in_bytes();
  const int count_offset  = java_lang_String::count_offset_in_bytes();
  const int offset_offset = java_lang_String::offset_offset_in_bytes();

  Node* no_ctrl = NULL;

  ciInstanceKlass* klass = env()->String_klass();
856
  const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(klass);
857 858 859 860 861 862 863 864 865 866 867 868 869 870

  const TypeAryPtr* value_type =
        TypeAryPtr::make(TypePtr::NotNull,
                         TypeAry::make(TypeInt::CHAR,TypeInt::POS),
                         ciTypeArrayKlass::make(T_CHAR), true, 0);

  // Get start addr of string and substring
  Node* str1_valuea  = basic_plus_adr(str1, str1, value_offset);
  Node* str1_value   = make_load(no_ctrl, str1_valuea, value_type, T_OBJECT, string_type->add_offset(value_offset));
  Node* str1_offseta = basic_plus_adr(str1, str1, offset_offset);
  Node* str1_offset  = make_load(no_ctrl, str1_offseta, TypeInt::INT, T_INT, string_type->add_offset(offset_offset));
  Node* str1_start   = array_element_address(str1_value, str1_offset, T_CHAR);

  Node* str2_valuea  = basic_plus_adr(str2, str2, value_offset);
871
  Node* str2_value   = make_load(no_ctrl, str2_valuea, value_type, T_OBJECT, string_type->add_offset(value_offset));
872
  Node* str2_offseta = basic_plus_adr(str2, str2, offset_offset);
873
  Node* str2_offset  = make_load(no_ctrl, str2_offseta, TypeInt::INT, T_INT, string_type->add_offset(offset_offset));
874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900
  Node* str2_start   = array_element_address(str2_value, str2_offset, T_CHAR);

  Node* result = NULL;
  switch (opcode) {
  case Op_StrIndexOf:
    result = new (C, 6) StrIndexOfNode(control(), memory(TypeAryPtr::CHARS),
                                       str1_start, cnt1, str2_start, cnt2);
    break;
  case Op_StrComp:
    result = new (C, 6) StrCompNode(control(), memory(TypeAryPtr::CHARS),
                                    str1_start, cnt1, str2_start, cnt2);
    break;
  case Op_StrEquals:
    result = new (C, 5) StrEqualsNode(control(), memory(TypeAryPtr::CHARS),
                                      str1_start, str2_start, cnt1);
    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
901 902 903
//------------------------------inline_string_compareTo------------------------
bool LibraryCallKit::inline_string_compareTo() {

C
cfang 已提交
904 905
  if (!Matcher::has_match_rule(Op_StrComp)) return false;

D
duke 已提交
906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926
  const int value_offset = java_lang_String::value_offset_in_bytes();
  const int count_offset = java_lang_String::count_offset_in_bytes();
  const int offset_offset = java_lang_String::offset_offset_in_bytes();

  _sp += 2;
  Node *argument = pop();  // pop non-receiver first:  it was pushed second
  Node *receiver = pop();

  // Null check on self without removing any arguments.  The argument
  // null check technically happens in the wrong place, which can lead to
  // invalid stack traces when string compare is inlined into a method
  // which handles NullPointerExceptions.
  _sp += 2;
  receiver = do_null_check(receiver, T_OBJECT);
  argument = do_null_check(argument, T_OBJECT);
  _sp -= 2;
  if (stopped()) {
    return true;
  }

  ciInstanceKlass* klass = env()->String_klass();
927
  const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(klass);
928 929 930 931 932 933 934 935
  Node* no_ctrl = NULL;

  // Get counts for string and argument
  Node* receiver_cnta = basic_plus_adr(receiver, receiver, count_offset);
  Node* receiver_cnt  = make_load(no_ctrl, receiver_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));

  Node* argument_cnta = basic_plus_adr(argument, argument, count_offset);
  Node* argument_cnt  = make_load(no_ctrl, argument_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));
D
duke 已提交
936

937
  Node* compare = make_string_method_node(Op_StrComp, receiver, receiver_cnt, argument, argument_cnt);
D
duke 已提交
938 939 940 941
  push(compare);
  return true;
}

C
cfang 已提交
942 943 944 945 946 947 948 949 950
//------------------------------inline_string_equals------------------------
bool LibraryCallKit::inline_string_equals() {

  if (!Matcher::has_match_rule(Op_StrEquals)) return false;

  const int value_offset = java_lang_String::value_offset_in_bytes();
  const int count_offset = java_lang_String::count_offset_in_bytes();
  const int offset_offset = java_lang_String::offset_offset_in_bytes();

951 952
  int nargs = 2;
  _sp += nargs;
C
cfang 已提交
953 954 955 956 957 958 959
  Node* argument = pop();  // pop non-receiver first:  it was pushed second
  Node* receiver = pop();

  // Null check on self without removing any arguments.  The argument
  // null check technically happens in the wrong place, which can lead to
  // invalid stack traces when string compare is inlined into a method
  // which handles NullPointerExceptions.
960
  _sp += nargs;
C
cfang 已提交
961 962 963
  receiver = do_null_check(receiver, T_OBJECT);
  //should not do null check for argument for String.equals(), because spec
  //allows to specify NULL as argument.
964
  _sp -= nargs;
C
cfang 已提交
965 966 967 968 969

  if (stopped()) {
    return true;
  }

970 971 972 973 974 975 976 977 978 979 980 981 982 983 984
  // paths (plus control) merge
  RegionNode* region = new (C, 5) RegionNode(5);
  Node* phi = new (C, 5) PhiNode(region, TypeInt::BOOL);

  // does source == target string?
  Node* cmp = _gvn.transform(new (C, 3) CmpPNode(receiver, argument));
  Node* bol = _gvn.transform(new (C, 2) BoolNode(cmp, BoolTest::eq));

  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
985 986 987
  // get String klass for instanceOf
  ciInstanceKlass* klass = env()->String_klass();

988
  if (!stopped()) {
989
    _sp += nargs;          // gen_instanceof might do an uncommon trap
990
    Node* inst = gen_instanceof(argument, makecon(TypeKlassPtr::make(klass)));
991
    _sp -= nargs;
992 993
    Node* cmp  = _gvn.transform(new (C, 3) CmpINode(inst, intcon(1)));
    Node* bol  = _gvn.transform(new (C, 2) BoolNode(cmp, BoolTest::ne));
C
cfang 已提交
994

995 996
    Node* inst_false = generate_guard(bol, NULL, PROB_MIN);
    //instanceOf == true, fallthrough
C
cfang 已提交
997

998 999 1000 1001 1002
    if (inst_false != NULL) {
      phi->init_req(3, intcon(0));
      region->init_req(3, inst_false);
    }
  }
C
cfang 已提交
1003

1004
  const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(klass);
C
cfang 已提交
1005

1006 1007 1008 1009 1010
  Node* no_ctrl = NULL;
  Node* receiver_cnt;
  Node* argument_cnt;

  if (!stopped()) {
1011 1012
    // Properly cast the argument to String
    argument = _gvn.transform(new (C, 2) CheckCastPPNode(control(), argument, string_type));
1013 1014
    // This path is taken only when argument's type is String:NotNull.
    argument = cast_not_null(argument, false);
1015

1016 1017 1018 1019 1020
    // Get counts for string and argument
    Node* receiver_cnta = basic_plus_adr(receiver, receiver, count_offset);
    receiver_cnt  = make_load(no_ctrl, receiver_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));

    Node* argument_cnta = basic_plus_adr(argument, argument, count_offset);
1021
    argument_cnt  = make_load(no_ctrl, argument_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031

    // Check for receiver count != argument count
    Node* cmp = _gvn.transform( new(C, 3) CmpINode(receiver_cnt, argument_cnt) );
    Node* bol = _gvn.transform( new(C, 2) BoolNode(cmp, BoolTest::ne) );
    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
1032

1033
  // Check for count == 0 is done by mach node StrEquals.
C
cfang 已提交
1034

1035 1036 1037 1038 1039
  if (!stopped()) {
    Node* equals = make_string_method_node(Op_StrEquals, receiver, receiver_cnt, argument, argument_cnt);
    phi->init_req(1, equals);
    region->init_req(1, control());
  }
C
cfang 已提交
1040 1041 1042 1043 1044 1045 1046 1047 1048 1049

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

  push(_gvn.transform(phi));

  return true;
}

1050 1051 1052
//------------------------------inline_array_equals----------------------------
bool LibraryCallKit::inline_array_equals() {

1053 1054
  if (!Matcher::has_match_rule(Op_AryEq)) return false;

1055 1056 1057 1058 1059
  _sp += 2;
  Node *argument2 = pop();
  Node *argument1 = pop();

  Node* equals =
1060 1061
    _gvn.transform(new (C, 4) AryEqNode(control(), memory(TypeAryPtr::CHARS),
                                        argument1, argument2) );
1062 1063 1064 1065
  push(equals);
  return true;
}

D
duke 已提交
1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133
// 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);

1134 1135
  const int nargs = 2; // number of arguments to push back for uncommon trap in predicate

D
duke 已提交
1136 1137 1138 1139 1140
  const int value_offset  = java_lang_String::value_offset_in_bytes();
  const int count_offset  = java_lang_String::count_offset_in_bytes();
  const int offset_offset = java_lang_String::offset_offset_in_bytes();

  ciInstanceKlass* klass = env()->String_klass();
1141
  const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(klass);
D
duke 已提交
1142 1143 1144 1145 1146 1147 1148 1149 1150
  const TypeAryPtr*  source_type = TypeAryPtr::make(TypePtr::NotNull, TypeAry::make(TypeInt::CHAR,TypeInt::POS), ciTypeArrayKlass::make(T_CHAR), true, 0);

  Node* sourceOffseta = basic_plus_adr(string_object, string_object, offset_offset);
  Node* sourceOffset  = make_load(no_ctrl, sourceOffseta, TypeInt::INT, T_INT, string_type->add_offset(offset_offset));
  Node* sourceCounta  = basic_plus_adr(string_object, string_object, count_offset);
  Node* sourceCount   = make_load(no_ctrl, sourceCounta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));
  Node* sourcea       = basic_plus_adr(string_object, string_object, value_offset);
  Node* source        = make_load(no_ctrl, sourcea, source_type, T_OBJECT, string_type->add_offset(value_offset));

1151
  Node* target = _gvn.transform( makecon(TypeOopPtr::make_from_constant(target_array, true)) );
D
duke 已提交
1152 1153 1154 1155
  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);

1156
  IdealKit kit(this, false, true);
D
duke 已提交
1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167
#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);

1168
  IdealVariable rtn(kit), i(kit), j(kit); __ declarations_done();
D
duke 已提交
1169 1170 1171 1172
  Node* outer_loop = __ make_label(2 /* goto */);
  Node* return_    = __ make_label(1);

  __ set(rtn,__ ConI(-1));
1173
  __ loop(this, nargs, i, sourceOffset, BoolTest::lt, sourceEnd); {
D
duke 已提交
1174 1175 1176 1177
       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); {
1178
         __ loop(this, nargs, j, zero, BoolTest::lt, targetCountLess1); {
D
duke 已提交
1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205
              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_);

1206
  // Final sync IdealKit and GraphKit.
1207
  final_sync(kit);
D
duke 已提交
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
  Node* result = __ value(rtn);
#undef __
  C->set_has_loops(true);
  return result;
}

//------------------------------inline_string_indexOf------------------------
bool LibraryCallKit::inline_string_indexOf() {

  const int value_offset  = java_lang_String::value_offset_in_bytes();
  const int count_offset  = java_lang_String::count_offset_in_bytes();
  const int offset_offset = java_lang_String::offset_offset_in_bytes();

C
cfang 已提交
1221 1222 1223
  _sp += 2;
  Node *argument = pop();  // pop non-receiver first:  it was pushed second
  Node *receiver = pop();
D
duke 已提交
1224

C
cfang 已提交
1225
  Node* result;
1226 1227
  // Disable the use of pcmpestri until it can be guaranteed that
  // the load doesn't cross into the uncommited space.
1228
  if (Matcher::has_match_rule(Op_StrIndexOf) &&
C
cfang 已提交
1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244
      UseSSE42Intrinsics) {
    // Generate SSE4.2 version of indexOf
    // We currently only have match rules that use SSE4.2

    // Null check on self without removing any arguments.  The argument
    // null check technically happens in the wrong place, which can lead to
    // invalid stack traces when string compare is inlined into a method
    // which handles NullPointerExceptions.
    _sp += 2;
    receiver = do_null_check(receiver, T_OBJECT);
    argument = do_null_check(argument, T_OBJECT);
    _sp -= 2;

    if (stopped()) {
      return true;
    }
D
duke 已提交
1245

1246 1247 1248
    ciInstanceKlass* str_klass = env()->String_klass();
    const TypeOopPtr* string_type = TypeOopPtr::make_from_klass(str_klass);

1249
    // Make the merge point
1250 1251
    RegionNode* result_rgn = new (C, 4) RegionNode(4);
    Node*       result_phi = new (C, 4) PhiNode(result_rgn, TypeInt::INT);
1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269
    Node* no_ctrl  = NULL;

    // Get counts for string and substr
    Node* source_cnta = basic_plus_adr(receiver, receiver, count_offset);
    Node* source_cnt  = make_load(no_ctrl, source_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));

    Node* substr_cnta = basic_plus_adr(argument, argument, count_offset);
    Node* substr_cnt  = make_load(no_ctrl, substr_cnta, TypeInt::INT, T_INT, string_type->add_offset(count_offset));

    // Check for substr count > string count
    Node* cmp = _gvn.transform( new(C, 3) CmpINode(substr_cnt, source_cnt) );
    Node* bol = _gvn.transform( new(C, 2) BoolNode(cmp, BoolTest::gt) );
    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);
    }

1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280
    if (!stopped()) {
      // Check for substr count == 0
      cmp = _gvn.transform( new(C, 3) CmpINode(substr_cnt, intcon(0)) );
      bol = _gvn.transform( new(C, 2) BoolNode(cmp, BoolTest::eq) );
      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);
      }
    }

1281 1282 1283 1284 1285 1286 1287 1288 1289
    if (!stopped()) {
      result = make_string_method_node(Op_StrIndexOf, receiver, source_cnt, argument, substr_cnt);
      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);

1290 1291
  } else { // Use LibraryCallKit::string_indexOf
    // don't intrinsify if argument isn't a constant string.
C
cfang 已提交
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
    if (!argument->is_Con()) {
     return false;
    }
    const TypeOopPtr* str_type = _gvn.type(argument)->isa_oopptr();
    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");

    ciObject* v = str->field_value_by_offset(value_offset).as_object();
    int       o = str->field_value_by_offset(offset_offset).as_int();
    int       c = str->field_value_by_offset(count_offset).as_int();
    ciTypeArray* pat = v->as_type_array(); // pattern (argument) character array

    // 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
1317

C
cfang 已提交
1318 1319 1320 1321 1322 1323 1324 1325 1326
    // Null check on self without removing any arguments.  The argument
    // null check technically happens in the wrong place, which can lead to
    // invalid stack traces when string compare is inlined into a method
    // which handles NullPointerExceptions.
    _sp += 2;
    receiver = do_null_check(receiver, T_OBJECT);
    // No null check on the argument is needed since it's a constant String oop.
    _sp -= 2;
    if (stopped()) {
1327
      return true;
C
cfang 已提交
1328
    }
D
duke 已提交
1329

C
cfang 已提交
1330 1331 1332 1333 1334
    // The null string as a pattern always returns 0 (match at beginning of string)
    if (c == 0) {
      push(intcon(0));
      return true;
    }
D
duke 已提交
1335

C
cfang 已提交
1336 1337 1338 1339 1340 1341 1342
    // 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
1343
    }
C
cfang 已提交
1344 1345 1346 1347 1348 1349 1350 1351 1352 1353

    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
1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 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 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564
  }

  push(result);
  return true;
}

//--------------------------pop_math_arg--------------------------------
// Pop a double argument to a math function from the stack
// rounding it if necessary.
Node * LibraryCallKit::pop_math_arg() {
  Node *arg = pop_pair();
  if( Matcher::strict_fp_requires_explicit_rounding && UseSSE<=1 )
    arg = _gvn.transform( new (C, 2) RoundDoubleNode(0, arg) );
  return arg;
}

//------------------------------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) {
  _sp += arg_size();            // restore stack pointer
  Node* arg = pop_math_arg();
  Node* trig = NULL;

  switch (id) {
  case vmIntrinsics::_dsin:
    trig = _gvn.transform((Node*)new (C, 2) SinDNode(arg));
    break;
  case vmIntrinsics::_dcos:
    trig = _gvn.transform((Node*)new (C, 2) CosDNode(arg));
    break;
  case vmIntrinsics::_dtan:
    trig = _gvn.transform((Node*)new (C, 2) TanDNode(arg));
    break;
  default:
    assert(false, "bad intrinsic was passed in");
    return false;
  }

  // Rounding required?  Check for argument reduction!
  if( Matcher::strict_fp_requires_explicit_rounding ) {

    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
    RegionNode *r = new (C, 3) RegionNode(3);
    Node *phi = new (C, 3) PhiNode(r,Type::DOUBLE);

    // Flatten arg so we need only 1 test
    Node *abs = _gvn.transform(new (C, 2) AbsDNode(arg));
    // Node for PI/4 constant
    Node *pi4 = makecon(TypeD::make(pi_4));
    // Check PI/4 : abs(arg)
    Node *cmp = _gvn.transform(new (C, 3) CmpDNode(pi4,abs));
    // Check: If PI/4 < abs(arg) then go slow
    Node *bol = _gvn.transform( new (C, 2) BoolNode( cmp, BoolTest::lt ) );
    // 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
    phi->init_req(2,trig);

    // 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, "");
    Node* slow_result = _gvn.transform(new (C, 1) ProjNode(call,TypeFunc::Parms));
    r->init_req(1,control());
    phi->init_req(1,slow_result);

    // Post-merge
    set_control(_gvn.transform(r));
    record_for_igvn(r);
    trig = _gvn.transform(phi);

    C->set_has_split_ifs(true); // Has chance for split-if optimization
  }
  // Push result back on JVM stack
  push_pair(trig);
  return true;
}

//------------------------------inline_sqrt-------------------------------------
// Inline square root instruction, if possible.
bool LibraryCallKit::inline_sqrt(vmIntrinsics::ID id) {
  assert(id == vmIntrinsics::_dsqrt, "Not square root");
  _sp += arg_size();        // restore stack pointer
  push_pair(_gvn.transform(new (C, 2) SqrtDNode(0, pop_math_arg())));
  return true;
}

//------------------------------inline_abs-------------------------------------
// Inline absolute value instruction, if possible.
bool LibraryCallKit::inline_abs(vmIntrinsics::ID id) {
  assert(id == vmIntrinsics::_dabs, "Not absolute value");
  _sp += arg_size();        // restore stack pointer
  push_pair(_gvn.transform(new (C, 2) AbsDNode(pop_math_arg())));
  return true;
}

//------------------------------inline_exp-------------------------------------
// Inline exp instructions, if possible.  The Intel hardware only misses
// really odd corner cases (+/- Infinity).  Just uncommon-trap them.
bool LibraryCallKit::inline_exp(vmIntrinsics::ID id) {
  assert(id == vmIntrinsics::_dexp, "Not exp");

  // If this inlining ever returned NaN in the past, we do not intrinsify it
  // every again.  NaN results requires StrictMath.exp handling.
  if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;

  // Do not intrinsify on older platforms which lack cmove.
  if (ConditionalMoveLimit == 0)  return false;

  _sp += arg_size();        // restore stack pointer
  Node *x = pop_math_arg();
  Node *result = _gvn.transform(new (C, 2) ExpDNode(0,x));

  //-------------------
  //result=(result.isNaN())? StrictMath::exp():result;
  // Check: If isNaN() by checking result!=result? then go to Strict Math
  Node* cmpisnan = _gvn.transform(new (C, 3) CmpDNode(result,result));
  // Build the boolean node
  Node* bolisnum = _gvn.transform( new (C, 2) BoolNode(cmpisnan, BoolTest::eq) );

  { BuildCutout unless(this, bolisnum, PROB_STATIC_FREQUENT);
    // End the current control-flow path
    push_pair(x);
    // Math.exp intrinsic returned a NaN, which requires StrictMath.exp
    // to handle.  Recompile without intrinsifying Math.exp
    uncommon_trap(Deoptimization::Reason_intrinsic,
                  Deoptimization::Action_make_not_entrant);
  }

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

  push_pair(result);

  return true;
}

//------------------------------inline_pow-------------------------------------
// Inline power instructions, if possible.
bool LibraryCallKit::inline_pow(vmIntrinsics::ID id) {
  assert(id == vmIntrinsics::_dpow, "Not pow");

  // If this inlining ever returned NaN in the past, we do not intrinsify it
  // every again.  NaN results requires StrictMath.pow handling.
  if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;

  // Do not intrinsify on older platforms which lack cmove.
  if (ConditionalMoveLimit == 0)  return false;

  // Pseudocode for pow
  // if (x <= 0.0) {
  //   if ((double)((int)y)==y) { // if y is int
  //     result = ((1&(int)y)==0)?-DPow(abs(x), y):DPow(abs(x), y)
  //   } else {
  //     result = NaN;
  //   }
  // } else {
  //   result = DPow(x,y);
  // }
  // if (result != result)?  {
T
twisti 已提交
1565
  //   uncommon_trap();
D
duke 已提交
1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 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
  // }
  // return result;

  _sp += arg_size();        // restore stack pointer
  Node* y = pop_math_arg();
  Node* x = pop_math_arg();

  Node *fast_result = _gvn.transform( new (C, 3) PowDNode(0, x, y) );

  // Short form: if not top-level (i.e., Math.pow but inlining Math.pow
  // inside of something) then skip the fancy tests and just check for
  // NaN result.
  Node *result = NULL;
  if( jvms()->depth() >= 1 ) {
    result = fast_result;
  } else {

    // Set the merge point for If node with condition of (x <= 0.0)
    // There are four possible paths to region node and phi node
    RegionNode *r = new (C, 4) RegionNode(4);
    Node *phi = new (C, 4) PhiNode(r, Type::DOUBLE);

    // Build the first if node: if (x <= 0.0)
    // Node for 0 constant
    Node *zeronode = makecon(TypeD::ZERO);
    // Check x:0
    Node *cmp = _gvn.transform(new (C, 3) CmpDNode(x, zeronode));
    // Check: If (x<=0) then go complex path
    Node *bol1 = _gvn.transform( new (C, 2) BoolNode( cmp, BoolTest::le ) );
    // Branch either way
    IfNode *if1 = create_and_xform_if(control(),bol1, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
    Node *opt_test = _gvn.transform(if1);
    //assert( opt_test->is_If(), "Expect an IfNode");
    IfNode *opt_if1 = (IfNode*)opt_test;
    // Fast path taken; set region slot 3
    Node *fast_taken = _gvn.transform( new (C, 1) IfFalseNode(opt_if1) );
    r->init_req(3,fast_taken); // Capture fast-control

    // Fast path not-taken, i.e. slow path
    Node *complex_path = _gvn.transform( new (C, 1) IfTrueNode(opt_if1) );

    // Set fast path result
    Node *fast_result = _gvn.transform( new (C, 3) PowDNode(0, y, x) );
    phi->init_req(3, fast_result);

    // Complex path
    // Build the second if node (if y is int)
    // Node for (int)y
    Node *inty = _gvn.transform( new (C, 2) ConvD2INode(y));
    // Node for (double)((int) y)
    Node *doubleinty= _gvn.transform( new (C, 2) ConvI2DNode(inty));
    // Check (double)((int) y) : y
    Node *cmpinty= _gvn.transform(new (C, 3) CmpDNode(doubleinty, y));
    // Check if (y isn't int) then go to slow path

    Node *bol2 = _gvn.transform( new (C, 2) BoolNode( cmpinty, BoolTest::ne ) );
T
twisti 已提交
1622
    // Branch either way
D
duke 已提交
1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
    IfNode *if2 = create_and_xform_if(complex_path,bol2, PROB_STATIC_INFREQUENT, COUNT_UNKNOWN);
    Node *slow_path = opt_iff(r,if2); // Set region path 2

    // Calculate DPow(abs(x), y)*(1 & (int)y)
    // Node for constant 1
    Node *conone = intcon(1);
    // 1& (int)y
    Node *signnode= _gvn.transform( new (C, 3) AndINode(conone, inty) );
    // zero node
    Node *conzero = intcon(0);
    // Check (1&(int)y)==0?
    Node *cmpeq1 = _gvn.transform(new (C, 3) CmpINode(signnode, conzero));
    // Check if (1&(int)y)!=0?, if so the result is negative
    Node *bol3 = _gvn.transform( new (C, 2) BoolNode( cmpeq1, BoolTest::ne ) );
    // abs(x)
    Node *absx=_gvn.transform( new (C, 2) AbsDNode(x));
    // abs(x)^y
    Node *absxpowy = _gvn.transform( new (C, 3) PowDNode(0, y, absx) );
    // -abs(x)^y
    Node *negabsxpowy = _gvn.transform(new (C, 2) NegDNode (absxpowy));
    // (1&(int)y)==1?-DPow(abs(x), y):DPow(abs(x), y)
    Node *signresult = _gvn.transform( CMoveNode::make(C, NULL, bol3, absxpowy, negabsxpowy, Type::DOUBLE));
    // Set complex path fast result
    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);
    result=_gvn.transform(phi);
  }

  //-------------------
  //result=(result.isNaN())? uncommon_trap():result;
  // Check: If isNaN() by checking result!=result? then go to Strict Math
  Node* cmpisnan = _gvn.transform(new (C, 3) CmpDNode(result,result));
  // Build the boolean node
  Node* bolisnum = _gvn.transform( new (C, 2) BoolNode(cmpisnan, BoolTest::eq) );

  { BuildCutout unless(this, bolisnum, PROB_STATIC_FREQUENT);
    // End the current control-flow path
    push_pair(x);
    push_pair(y);
    // Math.pow intrinsic returned a NaN, which requires StrictMath.pow
    // to handle.  Recompile without intrinsifying Math.pow.
    uncommon_trap(Deoptimization::Reason_intrinsic,
                  Deoptimization::Action_make_not_entrant);
  }

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

  push_pair(result);

  return true;
}

//------------------------------inline_trans-------------------------------------
// Inline transcendental instructions, if possible.  The Intel hardware gets
// these right, no funny corner cases missed.
bool LibraryCallKit::inline_trans(vmIntrinsics::ID id) {
  _sp += arg_size();        // restore stack pointer
  Node* arg = pop_math_arg();
  Node* trans = NULL;

  switch (id) {
  case vmIntrinsics::_dlog:
    trans = _gvn.transform((Node*)new (C, 2) LogDNode(arg));
    break;
  case vmIntrinsics::_dlog10:
    trans = _gvn.transform((Node*)new (C, 2) Log10DNode(arg));
    break;
  default:
    assert(false, "bad intrinsic was passed in");
    return false;
  }

  // Push result back on JVM stack
  push_pair(trans);
  return true;
}

//------------------------------runtime_math-----------------------------
bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
  Node* a = NULL;
  Node* b = NULL;

  assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
         "must be (DD)D or (D)D type");

  // Inputs
  _sp += arg_size();        // restore stack pointer
  if (call_type == OptoRuntime::Math_DD_D_Type()) {
    b = pop_math_arg();
  }
  a = pop_math_arg();

  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);
  Node* value = _gvn.transform(new (C, 1) ProjNode(trig, TypeFunc::Parms+0));
#ifdef ASSERT
  Node* value_top = _gvn.transform(new (C, 1) ProjNode(trig, TypeFunc::Parms+1));
  assert(value_top == top(), "second value must be top");
#endif

  push_pair(value);
  return true;
}

//------------------------------inline_math_native-----------------------------
bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
  switch (id) {
    // These intrinsics are not properly supported on all hardware
  case vmIntrinsics::_dcos: return Matcher::has_match_rule(Op_CosD) ? inline_trig(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
  case vmIntrinsics::_dsin: return Matcher::has_match_rule(Op_SinD) ? inline_trig(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
  case vmIntrinsics::_dtan: return Matcher::has_match_rule(Op_TanD) ? inline_trig(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");

  case vmIntrinsics::_dlog:   return Matcher::has_match_rule(Op_LogD) ? inline_trans(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
  case vmIntrinsics::_dlog10: return Matcher::has_match_rule(Op_Log10D) ? inline_trans(id) :
    runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");

    // These intrinsics are supported on all hardware
  case vmIntrinsics::_dsqrt: return Matcher::has_match_rule(Op_SqrtD) ? inline_sqrt(id) : false;
  case vmIntrinsics::_dabs:  return Matcher::has_match_rule(Op_AbsD)  ? inline_abs(id)  : false;

    // These intrinsics don't work on X86.  The ad implementation doesn't
    // handle NaN's properly.  Instead of returning infinity, the ad
    // implementation returns a NaN on overflow. See bug: 6304089
    // Once the ad implementations are fixed, change the code below
    // to match the intrinsics above

  case vmIntrinsics::_dexp:  return
    runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
  case vmIntrinsics::_dpow:  return
    runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");

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

  default:
    ShouldNotReachHere();
    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) {
  push(generate_min_max(id, argument(0), argument(1)));

  return true;
}

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;
  Node* ideal_cmpxy = _gvn.transform( new(C, 3) CmpINode(xkey, ykey) );
  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;
    best_bol = _gvn.transform( new(C, 2) BoolNode(cmpxy, BoolTest::lt) );
    // 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.
    base   = _gvn.transform( new (C, 2) CastX2PNode(offset) );
    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);
  }
}

2012 2013 2014 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 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053
//-------------------inline_numberOfLeadingZeros_int/long-----------------------
// inline int Integer.numberOfLeadingZeros(int)
// inline int Long.numberOfLeadingZeros(long)
bool LibraryCallKit::inline_numberOfLeadingZeros(vmIntrinsics::ID id) {
  assert(id == vmIntrinsics::_numberOfLeadingZeros_i || id == vmIntrinsics::_numberOfLeadingZeros_l, "not numberOfLeadingZeros");
  if (id == vmIntrinsics::_numberOfLeadingZeros_i && !Matcher::match_rule_supported(Op_CountLeadingZerosI)) return false;
  if (id == vmIntrinsics::_numberOfLeadingZeros_l && !Matcher::match_rule_supported(Op_CountLeadingZerosL)) return false;
  _sp += arg_size();  // restore stack pointer
  switch (id) {
  case vmIntrinsics::_numberOfLeadingZeros_i:
    push(_gvn.transform(new (C, 2) CountLeadingZerosINode(pop())));
    break;
  case vmIntrinsics::_numberOfLeadingZeros_l:
    push(_gvn.transform(new (C, 2) CountLeadingZerosLNode(pop_pair())));
    break;
  default:
    ShouldNotReachHere();
  }
  return true;
}

//-------------------inline_numberOfTrailingZeros_int/long----------------------
// inline int Integer.numberOfTrailingZeros(int)
// inline int Long.numberOfTrailingZeros(long)
bool LibraryCallKit::inline_numberOfTrailingZeros(vmIntrinsics::ID id) {
  assert(id == vmIntrinsics::_numberOfTrailingZeros_i || id == vmIntrinsics::_numberOfTrailingZeros_l, "not numberOfTrailingZeros");
  if (id == vmIntrinsics::_numberOfTrailingZeros_i && !Matcher::match_rule_supported(Op_CountTrailingZerosI)) return false;
  if (id == vmIntrinsics::_numberOfTrailingZeros_l && !Matcher::match_rule_supported(Op_CountTrailingZerosL)) return false;
  _sp += arg_size();  // restore stack pointer
  switch (id) {
  case vmIntrinsics::_numberOfTrailingZeros_i:
    push(_gvn.transform(new (C, 2) CountTrailingZerosINode(pop())));
    break;
  case vmIntrinsics::_numberOfTrailingZeros_l:
    push(_gvn.transform(new (C, 2) CountTrailingZerosLNode(pop_pair())));
    break;
  default:
    ShouldNotReachHere();
  }
  return true;
}

2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074
//----------------------------inline_bitCount_int/long-----------------------
// inline int Integer.bitCount(int)
// inline int Long.bitCount(long)
bool LibraryCallKit::inline_bitCount(vmIntrinsics::ID id) {
  assert(id == vmIntrinsics::_bitCount_i || id == vmIntrinsics::_bitCount_l, "not bitCount");
  if (id == vmIntrinsics::_bitCount_i && !Matcher::has_match_rule(Op_PopCountI)) return false;
  if (id == vmIntrinsics::_bitCount_l && !Matcher::has_match_rule(Op_PopCountL)) return false;
  _sp += arg_size();  // restore stack pointer
  switch (id) {
  case vmIntrinsics::_bitCount_i:
    push(_gvn.transform(new (C, 2) PopCountINode(pop())));
    break;
  case vmIntrinsics::_bitCount_l:
    push(_gvn.transform(new (C, 2) PopCountLNode(pop_pair())));
    break;
  default:
    ShouldNotReachHere();
  }
  return true;
}

2075
//----------------------------inline_reverseBytes_int/long/char/short-------------------
T
twisti 已提交
2076 2077
// inline Integer.reverseBytes(int)
// inline Long.reverseBytes(long)
2078 2079
// inline Character.reverseBytes(char)
// inline Short.reverseBytes(short)
D
duke 已提交
2080
bool LibraryCallKit::inline_reverseBytes(vmIntrinsics::ID id) {
2081 2082 2083 2084 2085 2086 2087
  assert(id == vmIntrinsics::_reverseBytes_i || id == vmIntrinsics::_reverseBytes_l ||
         id == vmIntrinsics::_reverseBytes_c || id == vmIntrinsics::_reverseBytes_s,
         "not reverse Bytes");
  if (id == vmIntrinsics::_reverseBytes_i && !Matcher::has_match_rule(Op_ReverseBytesI))  return false;
  if (id == vmIntrinsics::_reverseBytes_l && !Matcher::has_match_rule(Op_ReverseBytesL))  return false;
  if (id == vmIntrinsics::_reverseBytes_c && !Matcher::has_match_rule(Op_ReverseBytesUS)) return false;
  if (id == vmIntrinsics::_reverseBytes_s && !Matcher::has_match_rule(Op_ReverseBytesS))  return false;
D
duke 已提交
2088 2089 2090 2091 2092 2093 2094 2095
  _sp += arg_size();        // restore stack pointer
  switch (id) {
  case vmIntrinsics::_reverseBytes_i:
    push(_gvn.transform(new (C, 2) ReverseBytesINode(0, pop())));
    break;
  case vmIntrinsics::_reverseBytes_l:
    push_pair(_gvn.transform(new (C, 2) ReverseBytesLNode(0, pop_pair())));
    break;
2096 2097 2098 2099 2100 2101
  case vmIntrinsics::_reverseBytes_c:
    push(_gvn.transform(new (C, 2) ReverseBytesUSNode(0, pop())));
    break;
  case vmIntrinsics::_reverseBytes_s:
    push(_gvn.transform(new (C, 2) ReverseBytesSNode(0, pop())));
    break;
D
duke 已提交
2102 2103 2104 2105 2106 2107 2108 2109 2110 2111
  default:
    ;
  }
  return true;
}

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

const static BasicType T_ADDRESS_HOLDER = T_LONG;

2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 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
// Helper that guards and inserts a G1 pre-barrier.
void LibraryCallKit::insert_g1_pre_barrier(Node* base_oop, Node* offset, Node* pre_val) {
  assert(UseG1GC, "should not call this otherwise");

  // 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.

  // 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) {
      // Can the klass of base_oop be statically determined
      // to be _not_ a sub-class of Reference?
      ciKlass* klass = itype->klass();
      if (klass->is_subtype_of(env()->Reference_klass()) &&
          !env()->Reference_klass()->is_subtype_of(klass)) {
        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) {
  //   if (base != null) {
  //     if (klass(base)->reference_type() != REF_NONE)) {
  //       pre_barrier(_, pre_val, ...);
  //     }
  //   }
  // }

  float likely  = PROB_LIKELY(0.999);
  float unlikely  = PROB_UNLIKELY(0.999);

J
Merge  
johnc 已提交
2165
  IdealKit ideal(this);
2166 2167 2168 2169 2170
#define __ ideal.

  const int reference_type_offset = instanceKlass::reference_type_offset_in_bytes() +
                                        sizeof(oopDesc);

2171
  Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);
2172 2173 2174 2175 2176

  __ if_then(offset, BoolTest::eq, referent_off, unlikely); {
    __ if_then(base_oop, BoolTest::ne, null(), likely); {

      // Update graphKit memory and control from IdealKit.
J
Merge  
johnc 已提交
2177
      sync_kit(ideal);
2178 2179 2180 2181 2182

      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
2183
      __ sync_kit(this);
2184 2185 2186 2187 2188 2189

      Node* one = __ ConI(1);

      __ if_then(is_instof, BoolTest::eq, one, unlikely); {

        // Update graphKit from IdeakKit.
J
Merge  
johnc 已提交
2190
        sync_kit(ideal);
2191 2192 2193 2194

        // Use the pre-barrier to record the value in the referent field
        pre_barrier(false /* do_load */,
                    __ ctrl(),
2195
                    NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
2196 2197 2198 2199
                    pre_val /* pre_val */,
                    T_OBJECT);

        // Update IdealKit from graphKit.
J
Merge  
johnc 已提交
2200
        __ sync_kit(this);
2201 2202 2203 2204 2205 2206

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

  // Final sync IdealKit and GraphKit.
J
Merge  
johnc 已提交
2207
  final_sync(ideal);
2208 2209 2210 2211
#undef __
}


D
duke 已提交
2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 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
// Interpret Unsafe.fieldOffset cookies correctly:
extern jlong Unsafe_field_offset_to_byte_offset(jlong field_offset);

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.
    ciSignature* sig = signature();
#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".

  int type_words = type2size[ (type == T_ADDRESS) ? T_LONG : type ];

  // Argument words:  "this" plus (oop/offset) or (lo/hi) args plus maybe 1 or 2 value words
  int nargs = 1 + (is_native_ptr ? 2 : 3) + (is_store ? type_words : 0);

  debug_only(int saved_sp = _sp);
  _sp += nargs;

  Node* val;
  debug_only(val = (Node*)(uintptr_t)-1);


  if (is_store) {
    // Get the value being stored.  (Pop it first; it was pushed last.)
    switch (type) {
    case T_DOUBLE:
    case T_LONG:
    case T_ADDRESS:
      val = pop_pair();
      break;
    default:
      val = pop();
    }
  }

  // Build address expression.  See the code in inline_unsafe_prefetch.
  Node *adr;
  Node *heap_base_oop = top();
2288 2289
  Node* offset = top();

D
duke 已提交
2290 2291
  if (!is_native_ptr) {
    // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2292
    offset = pop_pair();
D
duke 已提交
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 2324 2325 2326 2327 2328 2329 2330 2331 2332
    // The base is either a Java object or a value produced by Unsafe.staticFieldBase
    Node* base   = pop();
    // 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;
  } else {
    Node* ptr = pop_pair();
    // Adjust Java long to machine word:
    ptr = ConvL2X(ptr);
    adr = make_unsafe_address(NULL, ptr);
  }

  // Pop receiver last:  it was pushed first.
  Node *receiver = pop();

  assert(saved_sp == _sp, "must have correct argument count");

  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);

2333 2334 2335 2336 2337 2338 2339
  // 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.
  bool need_read_barrier = UseG1GC && !is_native_ptr && !is_store &&
                           offset != top() && heap_base_oop != top();

D
duke 已提交
2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353
  if (!is_store && type == T_OBJECT) {
    // 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()) {
2354
      if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
D
duke 已提交
2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396
        const TypeOopPtr *elem_type = adr_type->is_aryptr()->elem()->isa_oopptr();
        if (elem_type != NULL) {
          sharpened_klass = elem_type->klass();
        }
      }
    }

    if (sharpened_klass != NULL) {
      const TypeOopPtr* tjp = TypeOopPtr::make_from_klass(sharpened_klass);

      // Sharpen the value type.
      value_type = tjp;

#ifndef PRODUCT
      if (PrintIntrinsics || PrintInlining || PrintOptoInlining) {
        tty->print("  from base type:  ");   adr_type->dump();
        tty->print("  sharpened value: "); value_type->dump();
      }
#endif
    }
  }

  // Null check on self without removing any arguments.  The argument
  // null check technically happens in the wrong place, which can lead to
  // invalid stack traces when the primitive is inlined into a method
  // which handles NullPointerExceptions.
  _sp += nargs;
  do_null_check(receiver, T_OBJECT);
  _sp -= nargs;
  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
2397
    // we cannot do effectively here because we probably only have a
D
duke 已提交
2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422
    // 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);
    // load value and push onto stack
    switch (type) {
    case T_BOOLEAN:
    case T_CHAR:
    case T_BYTE:
    case T_SHORT:
    case T_INT:
    case T_FLOAT:
2423 2424
      push(p);
      break;
D
duke 已提交
2425
    case T_OBJECT:
2426 2427 2428 2429
      if (need_read_barrier) {
        insert_g1_pre_barrier(heap_base_oop, offset, p);
      }
      push(p);
D
duke 已提交
2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461
      break;
    case T_ADDRESS:
      // Cast to an int type.
      p = _gvn.transform( new (C, 2) CastP2XNode(NULL,p) );
      p = ConvX2L(p);
      push_pair(p);
      break;
    case T_DOUBLE:
    case T_LONG:
      push_pair( p );
      break;
    default: ShouldNotReachHere();
    }
  } 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);
      val = _gvn.transform( new (C, 2) CastX2PNode(val) );
      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
2462
        (void) store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type);
D
duke 已提交
2463 2464 2465 2466 2467
      } 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.

2468
        IdealKit ideal(this);
2469
#define __ ideal.
D
duke 已提交
2470
        // QQQ who knows what probability is here??
2471 2472
        __ if_then(heap_base_oop, BoolTest::ne, null(), PROB_UNLIKELY(0.999)); {
          // Sync IdealKit and graphKit.
2473
          sync_kit(ideal);
2474 2475
          Node* st = store_oop_to_unknown(control(), heap_base_oop, adr, adr_type, val, type);
          // Update IdealKit memory.
2476
          __ sync_kit(this);
2477 2478 2479 2480
        } __ else_(); {
          __ store(__ ctrl(), adr, val, type, alias_type->index(), is_volatile);
        } __ end_if();
        // Final sync IdealKit and GraphKit.
2481
        final_sync(ideal);
2482
#undef __
D
duke 已提交
2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 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 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591
      }
    }
  }

  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.
    ciSignature* sig = signature();
#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".

  // Argument words:  "this" if not static, plus (oop/offset) or (lo/hi) args
  int nargs = (is_static ? 0 : 1) + (is_native_ptr ? 2 : 3);

  debug_only(int saved_sp = _sp);
  _sp += nargs;

  // Build address expression.  See the code in inline_unsafe_access.
  Node *adr;
  if (!is_native_ptr) {
    // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
    Node* offset = pop_pair();
    // The base is either a Java object or a value produced by Unsafe.staticFieldBase
    Node* base   = pop();
    // 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 {
    Node* ptr = pop_pair();
    // Adjust Java long to machine word:
    ptr = ConvL2X(ptr);
    adr = make_unsafe_address(NULL, ptr);
  }

  if (is_static) {
    assert(saved_sp == _sp, "must have correct argument count");
  } else {
    // Pop receiver last:  it was pushed first.
    Node *receiver = pop();
    assert(saved_sp == _sp, "must have correct argument count");

    // Null check on self without removing any arguments.  The argument
    // null check technically happens in the wrong place, which can lead to
    // invalid stack traces when the primitive is inlined into a method
    // which handles NullPointerExceptions.
    _sp += nargs;
    do_null_check(receiver, T_OBJECT);
    _sp -= nargs;
    if (stopped()) {
      return true;
    }
  }

  // Generate the read or write prefetch
  Node *prefetch;
  if (is_store) {
    prefetch = new (C, 3) PrefetchWriteNode(i_o(), adr);
  } else {
    prefetch = new (C, 3) PrefetchReadNode(i_o(), adr);
  }
  prefetch->init_req(0, control());
  set_i_o(_gvn.transform(prefetch));

  return true;
}

//----------------------------inline_unsafe_CAS----------------------------

bool LibraryCallKit::inline_unsafe_CAS(BasicType type) {
  // 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
2592
  // the correspondences clearer. - dl
D
duke 已提交
2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614

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

#ifndef PRODUCT
  {
    ResourceMark rm;
    // Check the signatures.
    ciSignature* sig = signature();
#ifdef ASSERT
    BasicType rtype = sig->return_type()->basic_type();
    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
  }
#endif //PRODUCT

  // number of stack slots per value argument (1 or 2)
  int type_words = type2size[type];

  // Cannot inline wide CAS on machines that don't support it natively
2615
  if (type2aelembytes(type) > BytesPerInt && !VM_Version::supports_cx8())
D
duke 已提交
2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658
    return false;

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

  // Argument words:  "this" plus oop plus offset plus oldvalue plus newvalue;
  int nargs = 1 + 1 + 2  + type_words + type_words;

  // pop arguments: newval, oldval, offset, base, and receiver
  debug_only(int saved_sp = _sp);
  _sp += nargs;
  Node* newval   = (type_words == 1) ? pop() : pop_pair();
  Node* oldval   = (type_words == 1) ? pop() : pop_pair();
  Node *offset   = pop_pair();
  Node *base     = pop();
  Node *receiver = pop();
  assert(saved_sp == _sp, "must have correct argument count");

  //  Null check receiver.
  _sp += nargs;
  do_null_check(receiver, T_OBJECT);
  _sp -= nargs;
  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();

  // (Unlike inline_unsafe_access, there seems no point in trying
  // to refine types. Just use the coarse types here.
  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");
  int alias_idx = C->get_alias_index(adr_type);

  // Memory-model-wise, a CAS acts like a little synchronized block,
T
twisti 已提交
2659
  // so needs barriers on each side.  These don't translate into
D
duke 已提交
2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681
  // actual barriers on most machines, but we still need rest of
  // 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.
  Node* cas;
  switch(type) {
  case T_INT:
    cas = _gvn.transform(new (C, 5) CompareAndSwapINode(control(), mem, adr, newval, oldval));
    break;
  case T_LONG:
    cas = _gvn.transform(new (C, 5) CompareAndSwapLNode(control(), mem, adr, newval, oldval));
    break;
  case T_OBJECT:
2682
     // reference stores need a store barrier.
D
duke 已提交
2683
    // (They don't if CAS fails, but it isn't worth checking.)
2684 2685 2686 2687
    pre_barrier(true /* do_load*/,
                control(), base, adr, alias_idx, newval, value_type->make_oopptr(),
                NULL /* pre_val*/,
                T_OBJECT);
2688
#ifdef _LP64
2689
    if (adr->bottom_type()->is_ptr_to_narrowoop()) {
2690 2691
      Node *newval_enc = _gvn.transform(new (C, 2) EncodePNode(newval, newval->bottom_type()->make_narrowoop()));
      Node *oldval_enc = _gvn.transform(new (C, 2) EncodePNode(oldval, oldval->bottom_type()->make_narrowoop()));
2692
      cas = _gvn.transform(new (C, 5) CompareAndSwapNNode(control(), mem, adr,
2693
                                                          newval_enc, oldval_enc));
2694 2695
    } else
#endif
2696 2697 2698
    {
      cas = _gvn.transform(new (C, 5) CompareAndSwapPNode(control(), mem, adr, newval, oldval));
    }
D
duke 已提交
2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781
    post_barrier(control(), cas, base, adr, alias_idx, newval, T_OBJECT, true);
    break;
  default:
    ShouldNotReachHere();
    break;
  }

  // SCMemProjNodes represent the memory state of CAS. Their main
  // role is to prevent CAS nodes from being optimized away when their
  // results aren't used.
  Node* proj = _gvn.transform( new (C, 1) SCMemProjNode(cas));
  set_memory(proj, alias_idx);

  // Add the trailing membar surrounding the access
  insert_mem_bar(Op_MemBarCPUOrder);
  insert_mem_bar(Op_MemBarAcquire);

  push(cas);
  return true;
}

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.
    ciSignature* sig = signature();
#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

  // number of stack slots per value argument (1 or 2)
  int type_words = type2size[type];

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

  // Argument words:  "this" plus oop plus offset plus value;
  int nargs = 1 + 1 + 2 + type_words;

  // pop arguments: val, offset, base, and receiver
  debug_only(int saved_sp = _sp);
  _sp += nargs;
  Node* val      = (type_words == 1) ? pop() : pop_pair();
  Node *offset   = pop_pair();
  Node *base     = pop();
  Node *receiver = pop();
  assert(saved_sp == _sp, "must have correct argument count");

  //  Null check receiver.
  _sp += nargs;
  do_null_check(receiver, T_OBJECT);
  _sp -= nargs;
  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:
  bool require_atomic_access = true;
  Node* store;
  if (type == T_OBJECT) // reference stores need a store barrier.
N
never 已提交
2782
    store = store_oop_to_unknown(control(), base, adr, adr_type, val, type);
D
duke 已提交
2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885
  else {
    store = store_to_memory(control(), adr, val, type, adr_type, require_atomic_access);
  }
  insert_mem_bar(Op_MemBarCPUOrder);
  return true;
}

bool LibraryCallKit::inline_unsafe_allocate() {
  if (callee()->is_static())  return false;  // caller must have the capability!
  int nargs = 1 + 1;
  assert(signature()->size() == nargs-1, "alloc has 1 argument");
  null_check_receiver(callee());  // check then ignore argument(0)
  _sp += nargs;  // set original stack for use by uncommon_trap
  Node* cls = do_null_check(argument(1), T_OBJECT);
  _sp -= nargs;
  if (stopped())  return true;

  Node* kls = load_klass_from_mirror(cls, false, nargs, NULL, 0);
  _sp += nargs;  // set original stack for use by uncommon_trap
  kls = do_null_check(kls, T_OBJECT);
  _sp -= nargs;
  if (stopped())  return true;  // argument was like int.class

  // 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, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc));
  Node* inst = make_load(NULL, insp, TypeInt::INT, T_INT);
  Node* bits = intcon(instanceKlass::fully_initialized);
  Node* test = _gvn.transform( new (C, 3) SubINode(inst, bits) );
  // The 'test' is non-zero if we need to take a slow path.

  Node* obj = new_instance(kls, test);
  push(obj);

  return true;
}

//------------------------inline_native_time_funcs--------------
// inline code for System.currentTimeMillis() and System.nanoTime()
// these have the same type and signature
bool LibraryCallKit::inline_native_time_funcs(bool isNano) {
  address funcAddr = isNano ? CAST_FROM_FN_PTR(address, os::javaTimeNanos) :
                              CAST_FROM_FN_PTR(address, os::javaTimeMillis);
  const char * funcName = isNano ? "nanoTime" : "currentTimeMillis";
  const TypeFunc *tf = OptoRuntime::current_time_millis_Type();
  const TypePtr* no_memory_effects = NULL;
  Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
  Node* value = _gvn.transform(new (C, 1) ProjNode(time, TypeFunc::Parms+0));
#ifdef ASSERT
  Node* value_top = _gvn.transform(new (C, 1) ProjNode(time, TypeFunc::Parms + 1));
  assert(value_top == top(), "second value must be top");
#endif
  push_pair(value);
  return true;
}

//------------------------inline_native_currentThread------------------
bool LibraryCallKit::inline_native_currentThread() {
  Node* junk = NULL;
  push(generate_current_thread(junk));
  return true;
}

//------------------------inline_native_isInterrupted------------------
bool LibraryCallKit::inline_native_isInterrupted() {
  const int nargs = 1+1;  // receiver + boolean
  assert(nargs == arg_size(), "sanity");
  // 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.
  RegionNode* slow_region = new (C, 1) RegionNode(1);
  record_for_igvn(slow_region);
  RegionNode* result_rgn = new (C, 4) RegionNode(1+3); // fast1, fast2, slow
  PhiNode*    result_val = new (C, 4) PhiNode(result_rgn, TypeInt::BOOL);
  enum { no_int_result_path   = 1,
         no_clear_result_path = 2,
         slow_result_path     = 3
  };

  // (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);
  Node* cmp_thr = _gvn.transform( new (C, 3) CmpPNode(cur_thr, rec_thr) );
  Node* bol_thr = _gvn.transform( new (C, 2) BoolNode(cmp_thr, BoolTest::ne) );

  bool known_current_thread = (_gvn.type(bol_thr) == TypeInt::ZERO);
  if (!known_current_thread)
    generate_slow_guard(bol_thr, slow_region);

  // (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()));
2886 2887
  // 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);
D
duke 已提交
2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976
  Node* cmp_bit = _gvn.transform( new (C, 3) CmpINode(int_bit, intcon(0)) );
  Node* bol_bit = _gvn.transform( new (C, 2) BoolNode(cmp_bit, BoolTest::ne) );

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

  // First fast path:  if (!TLS._interrupted) return false;
  Node* false_bit = _gvn.transform( new (C, 1) IfFalseNode(iff_bit) );
  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
  set_control( _gvn.transform(new (C, 1) IfTrueNode(iff_bit)) );

  // (c) Or, if interrupt bit is set and clear_int is false, use 2nd fast path.
  Node* clr_arg = argument(1);
  Node* cmp_arg = _gvn.transform( new (C, 3) CmpINode(clr_arg, intcon(0)) );
  Node* bol_arg = _gvn.transform( new (C, 2) BoolNode(cmp_arg, BoolTest::ne) );
  IfNode* iff_arg = create_and_map_if(control(), bol_arg, PROB_FAIR, COUNT_UNKNOWN);

  // Second fast path:  ... else if (!clear_int) return true;
  Node* false_arg = _gvn.transform( new (C, 1) IfFalseNode(iff_arg) );
  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
  set_control( _gvn.transform(new (C, 1) IfTrueNode(iff_arg)) );

  // (d) Otherwise, go to the slow path.
  slow_region->add_req(control());
  set_control( _gvn.transform(slow_region) );

  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

    // If we know that the result of the slow call will be true, tell the optimizer!
    if (known_current_thread)  slow_val = intcon(1);

    Node* fast_io  = slow_call->in(TypeFunc::I_O);
    Node* fast_mem = slow_call->in(TypeFunc::Memory);
    // These two phis are pre-filled with copies of of the fast IO and Memory
    Node* io_phi   = PhiNode::make(result_rgn, fast_io,  Type::ABIO);
    Node* mem_phi  = PhiNode::make(result_rgn, fast_mem, Type::MEMORY, TypePtr::BOTTOM);

    result_rgn->init_req(slow_result_path, control());
    io_phi    ->init_req(slow_result_path, i_o());
    mem_phi   ->init_req(slow_result_path, reset_memory());
    result_val->init_req(slow_result_path, slow_val);

    set_all_memory( _gvn.transform(mem_phi) );
    set_i_o(        _gvn.transform(io_phi) );
  }

  push_result(result_rgn, result_val);
  C->set_has_split_ifs(true); // Has chance for split-if optimization

  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) {
  Node* p = basic_plus_adr(klass, Klass::java_mirror_offset_in_bytes() + sizeof(oopDesc));
  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,
                                                    int nargs,
                                                    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;
2977
  Node* kls = _gvn.transform( LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type) );
D
duke 已提交
2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111
  _sp += nargs; // any deopt will start just before call to enclosing method
  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");
  }
  _sp -= nargs;
  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.
  Node* modp = basic_plus_adr(kls, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc));
  Node* mods = make_load(NULL, modp, TypeInt::INT, T_INT);
  Node* mask = intcon(modifier_mask);
  Node* bits = intcon(modifier_bits);
  Node* mbit = _gvn.transform( new (C, 3) AndINode(mods, mask) );
  Node* cmp  = _gvn.transform( new (C, 3) CmpINode(mbit, bits) );
  Node* bol  = _gvn.transform( new (C, 2) BoolNode(cmp, BoolTest::ne) );
  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) {
  int nargs = 1+0;  // just the Class mirror, in most cases
  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 };

  switch (id) {
  case vmIntrinsics::_isInstance:
    nargs = 1+1;  // the Class mirror, plus the object getting queried about
    // nothing is an instance of a primitive type
    prim_return_value = intcon(0);
    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:
    ShouldNotReachHere();
  }

  Node* mirror =                      argument(0);
  Node* obj    = (nargs <= 1)? top(): argument(1);

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

#ifndef PRODUCT
  if (PrintIntrinsics || PrintInlining || PrintOptoInlining) {
    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).
  RegionNode* region = new (C, PATH_LIMIT) RegionNode(PATH_LIMIT);
  record_for_igvn(region);
  PhiNode* phi = new (C, PATH_LIMIT) PhiNode(region, return_type);

  // 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.
  _sp += nargs;  // set original stack for use by uncommon_trap
  mirror = do_null_check(mirror, T_OBJECT);
  _sp -= nargs;
  // 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.
  Node* kls = load_klass_from_mirror(mirror, never_see_null, nargs,
                                     region, _prim_path);
  // If kls is null, we have a primitive mirror.
  phi->init_req(_prim_path, prim_return_value);
  if (stopped()) { push_result(region, phi); return true; }

  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
3112
    _sp += nargs;          // gen_instanceof might do an uncommon trap
D
duke 已提交
3113
    query_value = gen_instanceof(obj, kls);
3114
    _sp -= nargs;
D
duke 已提交
3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158
    break;

  case vmIntrinsics::_getModifiers:
    p = basic_plus_adr(kls, Klass::modifier_flags_offset_in_bytes() + sizeof(oopDesc));
    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.
    p = basic_plus_adr(kls, Klass::super_offset_in_bytes() + sizeof(oopDesc));
3159
    kls = _gvn.transform( LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeKlassPtr::OBJECT_OR_NULL) );
D
duke 已提交
3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244
    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);
      Node* cma = basic_plus_adr(kls, in_bytes(arrayKlass::component_mirror_offset()) + sizeof(oopDesc));
      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:
    p = basic_plus_adr(kls, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc));
    query_value = make_load(NULL, p, TypeInt::INT, T_INT);
    break;

  default:
    ShouldNotReachHere();
  }

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

  push_result(region, phi);
  C->set_has_split_ifs(true); // Has chance for split-if optimization

  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() {
  int nargs = 1+1;  // the Class mirror, plus the other class getting examined

  // 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
  };

  RegionNode* region = new (C, PATH_LIMIT) RegionNode(PATH_LIMIT);
  Node*       phi    = new (C, PATH_LIMIT) PhiNode(region, TypeInt::BOOL);
  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];
    _sp += nargs;  // set original stack for use by uncommon_trap
    arg = do_null_check(arg, T_OBJECT);
    _sp -= nargs;
    if (stopped())  break;
    args[which_arg] = _gvn.transform(arg);

    Node* p = basic_plus_adr(arg, class_klass_offset);
3245
    Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
D
duke 已提交
3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362
    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();
    _sp += nargs;  // set original stack for use by uncommon_trap
    kls = null_check_oop(kls, &null_ctl, never_see_null);
    _sp -= nargs;
    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.
    Node* cmp_eq = _gvn.transform( new (C, 3) CmpPNode(args[0], args[1]) );
    Node* bol_eq = _gvn.transform( new (C, 2) BoolNode(cmp_eq, BoolTest::eq) );
    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));
  push(_gvn.transform(phi));

  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)
                  : Klass::layout_helper_is_javaArray(layout_con));
    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);
  Node* cmp = _gvn.transform( new(C, 3) CmpINode(layout_val, intcon(nval)) );
  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();
  Node* bol = _gvn.transform( new(C, 2) BoolNode(cmp, btest) );
  return generate_fair_guard(bol, region);
}


//-----------------------inline_native_newArray--------------------------
bool LibraryCallKit::inline_native_newArray() {
  int nargs = 2;
  Node* mirror    = argument(0);
  Node* count_val = argument(1);

  _sp += nargs;  // set original stack for use by uncommon_trap
  mirror = do_null_check(mirror, T_OBJECT);
  _sp -= nargs;
3363 3364
  // If mirror or obj is dead, only null-path is taken.
  if (stopped())  return true;
D
duke 已提交
3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402

  enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
  RegionNode* result_reg = new(C, PATH_LIMIT) RegionNode(PATH_LIMIT);
  PhiNode*    result_val = new(C, PATH_LIMIT) PhiNode(result_reg,
                                                      TypeInstPtr::NOTNULL);
  PhiNode*    result_io  = new(C, PATH_LIMIT) PhiNode(result_reg, Type::ABIO);
  PhiNode*    result_mem = new(C, PATH_LIMIT) PhiNode(result_reg, Type::MEMORY,
                                                      TypePtr::BOTTOM);

  bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
  Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
                                                  nargs,
                                                  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.
3403
    Node* obj = new_array(klass_node, count_val, nargs);
D
duke 已提交
3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 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 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466
    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)  );
  set_all_memory( _gvn.transform(result_mem) );
  push_result(result_reg, result_val);
  C->set_has_split_ifs(true); // Has chance for split-if optimization

  return true;
}

//----------------------inline_native_getLength--------------------------
bool LibraryCallKit::inline_native_getLength() {
  if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;

  int nargs = 1;
  Node* array = argument(0);

  _sp += nargs;  // set original stack for use by uncommon_trap
  array = do_null_check(array, T_OBJECT);
  _sp -= nargs;

  // 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);
    _sp += nargs;  // push the arguments back on the stack
    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.
  push( load_array_length(array) );

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

  return true;
}

//------------------------inline_array_copyOf----------------------------
bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
  if (too_many_traps(Deoptimization::Reason_intrinsic))  return false;

  // Restore the stack and pop off the arguments.
  int nargs = 3 + (is_copyOfRange? 1: 0);
  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);

3467
  Node* newcopy;
D
duke 已提交
3468

3469 3470 3471 3472 3473
  //set the original stack and the reexecute bit for the interpreter to reexecute
  //the bytecode that invokes Arrays.copyOf if deoptimization happens
  { PreserveReexecuteState preexecs(this);
    _sp += nargs;
    jvms()->set_should_reexecute(true);
D
duke 已提交
3474

3475 3476
    array_type_mirror = do_null_check(array_type_mirror, T_OBJECT);
    original          = do_null_check(original, T_OBJECT);
D
duke 已提交
3477

3478 3479
    // Check if a null path was taken unconditionally.
    if (stopped())  return true;
D
duke 已提交
3480

3481
    Node* orig_length = load_array_length(original);
D
duke 已提交
3482

3483 3484 3485
    Node* klass_node = load_klass_from_mirror(array_type_mirror, false, 0,
                                              NULL, 0);
    klass_node = do_null_check(klass_node, T_OBJECT);
D
duke 已提交
3486

3487 3488
    RegionNode* bailout = new (C, 1) RegionNode(1);
    record_for_igvn(bailout);
D
duke 已提交
3489

3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500
    // 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*/);
      Node* cast = new (C, 2) CastPPNode(klass_node, akls);
      cast->init_req(0, control());
      klass_node = _gvn.transform(cast);
    }
D
duke 已提交
3501

3502 3503 3504
    // 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
3505

3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522
    Node* length = end;
    if (_gvn.type(start) != TypeInt::ZERO) {
      length = _gvn.transform( new (C, 3) SubINode(end, start) );
    }

    // Bail out if length is negative.
    // ...Not needed, since the new_array will throw the right exception.
    //generate_negative_guard(length, bailout, &length);

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

    if (!stopped()) {
3523 3524 3525 3526 3527 3528

      // How many elements will we copy from the original?
      // The answer is MinI(orig_length - start, length).
      Node* orig_tail = _gvn.transform( new(C, 3) SubINode(orig_length, start) );
      Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);

3529
      newcopy = new_array(klass_node, length, 0);
3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540

      // 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;
      bool length_never_negative = true;
      generate_arraycopy(TypeAryPtr::OOPS, T_OBJECT,
                         original, start, newcopy, intcon(0), moved,
                         disjoint_bases, length_never_negative);
3541 3542
    }
  } //original reexecute and sp are set back here
D
duke 已提交
3543

3544
  if(!stopped()) {
D
duke 已提交
3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 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 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 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 3657 3658 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 3685 3686 3687 3688 3689
    push(newcopy);
  }

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

  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();
  // Get the methodOop out of the appropriate vtable entry.
  int entry_offset  = (instanceKlass::vtable_start_offset() +
                     vtable_index*vtableEntry::size()) * wordSize +
                     vtableEntry::method_offset_in_bytes();
  Node* entry_addr  = basic_plus_adr(obj_klass, entry_offset);
  Node* target_call = make_load(NULL, entry_addr, TypeInstPtr::NOTNULL, T_OBJECT);

  // Compare the target method with the expected method (e.g., Object.hashCode).
  const TypeInstPtr* native_call_addr = TypeInstPtr::make(method);

  Node* native_call = makecon(native_call_addr);
  Node* chk_native  = _gvn.transform( new(C, 3) CmpPNode(target_call, native_call) );
  Node* test_native = _gvn.transform( new(C, 2) BoolNode(chk_native, BoolTest::ne) );

  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);
  int tfdc = tf->domain()->cnt();
  CallJavaNode* slow_call;
  if (is_static) {
    assert(!is_virtual, "");
    slow_call = new(C, tfdc) CallStaticJavaNode(tf,
                                SharedRuntime::get_resolve_static_call_stub(),
                                method, bci());
  } else if (is_virtual) {
    null_check_receiver(method);
    int vtable_index = methodOopDesc::invalid_vtable_index;
    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();
    }
    slow_call = new(C, tfdc) CallDynamicJavaNode(tf,
                                SharedRuntime::get_resolve_virtual_call_stub(),
                                method, vtable_index, bci());
  } else {  // neither virtual nor static:  opt_virtual
    null_check_receiver(method);
    slow_call = new(C, tfdc) CallStaticJavaNode(tf,
                                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 };

  RegionNode* result_reg = new(C, PATH_LIMIT) RegionNode(PATH_LIMIT);
  PhiNode*    result_val = new(C, PATH_LIMIT) PhiNode(result_reg,
                                                      TypeInt::INT);
  PhiNode*    result_io  = new(C, PATH_LIMIT) PhiNode(result_reg, Type::ABIO);
  PhiNode*    result_mem = new(C, PATH_LIMIT) PhiNode(result_reg, Type::MEMORY,
                                                      TypePtr::BOTTOM);
  Node* obj = NULL;
  if (!is_static) {
    // Check for hashing null object
    obj = null_check_receiver(callee());
    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()) {
    set_control( result_reg->in(_null_path) );
    if (!stopped())
      push(      result_val ->in(_null_path) );
    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.
  RegionNode* slow_region = new (C, 1) RegionNode(1);
  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());
3690
  Node* header = make_load(control(), header_addr, TypeX_X, TypeX_X->basic_type());
D
duke 已提交
3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708

  // Test the header to see if it is unlocked.
  Node *lock_mask      = _gvn.MakeConX(markOopDesc::biased_lock_mask_in_place);
  Node *lmasked_header = _gvn.transform( new (C, 3) AndXNode(header, lock_mask) );
  Node *unlocked_val   = _gvn.MakeConX(markOopDesc::unlocked_value);
  Node *chk_unlocked   = _gvn.transform( new (C, 3) CmpXNode( lmasked_header, unlocked_val));
  Node *test_unlocked  = _gvn.transform( new (C, 2) BoolNode( chk_unlocked, BoolTest::ne) );

  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);
  Node *hshifted_header= _gvn.transform( new (C, 3) URShiftXNode(header, hash_shift) );
  // This hack lets the hash bits live anywhere in the mark object now, as long
T
twisti 已提交
3709
  // as the shift drops the relevant bits into the low 32 bits.  Note that
D
duke 已提交
3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755
  // 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);
  Node *hash_val       = _gvn.transform( new (C, 3) AndINode(hshifted_header, hash_mask) );

  Node *no_hash_val    = _gvn.intcon(markOopDesc::no_hash);
  Node *chk_assigned   = _gvn.transform( new (C, 3) CmpINode( hash_val, no_hash_val));
  Node *test_assigned  = _gvn.transform( new (C, 2) BoolNode( chk_assigned, BoolTest::eq) );

  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);
    vmIntrinsics::ID hashCode_id = vmIntrinsics::_hashCode;
    if (is_static)   hashCode_id = vmIntrinsics::_identityHashCode;
    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)  );
  set_all_memory( _gvn.transform(result_mem) );
  push_result(result_reg, result_val);

  return true;
}

//---------------------------inline_native_getClass----------------------------
T
twisti 已提交
3756
// Build special case code for calls to getClass on an object.
D
duke 已提交
3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888
bool LibraryCallKit::inline_native_getClass() {
  Node* obj = null_check_receiver(callee());
  if (stopped())  return true;
  push( load_mirror_from_klass(load_object_klass(obj)) );
  return true;
}

//-----------------inline_native_Reflection_getCallerClass---------------------
// In the presence of deep enough inlining, getCallerClass() becomes a no-op.
//
// NOTE that this code must perform the same logic as
// vframeStream::security_get_caller_frame in that it must skip
// Method.invoke() and auxiliary frames.




bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
  ciMethod*       method = callee();

#ifndef PRODUCT
  if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) {
    tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
  }
#endif

  debug_only(int saved_sp = _sp);

  // Argument words:  (int depth)
  int nargs = 1;

  _sp += nargs;
  Node* caller_depth_node = pop();

  assert(saved_sp == _sp, "must have correct argument count");

  // The depth value must be a constant in order for the runtime call
  // to be eliminated.
  const TypeInt* caller_depth_type = _gvn.type(caller_depth_node)->isa_int();
  if (caller_depth_type == NULL || !caller_depth_type->is_con()) {
#ifndef PRODUCT
    if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) {
      tty->print_cr("  Bailing out because caller depth was not a constant");
    }
#endif
    return false;
  }
  // Note that the JVM state at this point does not include the
  // getCallerClass() frame which we are trying to inline. The
  // semantics of getCallerClass(), however, are that the "first"
  // frame is the getCallerClass() frame, so we subtract one from the
  // requested depth before continuing. We don't inline requests of
  // getCallerClass(0).
  int caller_depth = caller_depth_type->get_con() - 1;
  if (caller_depth < 0) {
#ifndef PRODUCT
    if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) {
      tty->print_cr("  Bailing out because caller depth was %d", caller_depth);
    }
#endif
    return false;
  }

  if (!jvms()->has_method()) {
#ifndef PRODUCT
    if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) {
      tty->print_cr("  Bailing out because intrinsic was inlined at top level");
    }
#endif
    return false;
  }
  int _depth = jvms()->depth();  // cache call chain depth

  // Walk back up the JVM state to find the caller at the required
  // depth. NOTE that this code must perform the same logic as
  // vframeStream::security_get_caller_frame in that it must skip
  // Method.invoke() and auxiliary frames. Note also that depth is
  // 1-based (1 is the bottom of the inlining).
  int inlining_depth = _depth;
  JVMState* caller_jvms = NULL;

  if (inlining_depth > 0) {
    caller_jvms = jvms();
    assert(caller_jvms = jvms()->of_depth(inlining_depth), "inlining_depth == our depth");
    do {
      // The following if-tests should be performed in this order
      if (is_method_invoke_or_aux_frame(caller_jvms)) {
        // Skip a Method.invoke() or auxiliary frame
      } else if (caller_depth > 0) {
        // Skip real frame
        --caller_depth;
      } else {
        // We're done: reached desired caller after skipping.
        break;
      }
      caller_jvms = caller_jvms->caller();
      --inlining_depth;
    } while (inlining_depth > 0);
  }

  if (inlining_depth == 0) {
#ifndef PRODUCT
    if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) {
      tty->print_cr("  Bailing out because caller depth (%d) exceeded inlining depth (%d)", caller_depth_type->get_con(), _depth);
      tty->print_cr("  JVM state at this point:");
      for (int i = _depth; i >= 1; i--) {
        tty->print_cr("   %d) %s", i, jvms()->of_depth(i)->method()->name()->as_utf8());
      }
    }
#endif
    return false; // Reached end of inlining
  }

  // Acquire method holder as java.lang.Class
  ciInstanceKlass* caller_klass  = caller_jvms->method()->holder();
  ciInstance*      caller_mirror = caller_klass->java_mirror();
  // Push this as a constant
  push(makecon(TypeInstPtr::make(caller_mirror)));
#ifndef PRODUCT
  if ((PrintIntrinsics || PrintInlining || PrintOptoInlining) && Verbose) {
    tty->print_cr("  Succeeded: caller = %s.%s, caller depth = %d, depth = %d", caller_klass->name()->as_utf8(), caller_jvms->method()->name()->as_utf8(), caller_depth_type->get_con(), _depth);
    tty->print_cr("  JVM state at this point:");
    for (int i = _depth; i >= 1; i--) {
      tty->print_cr("   %d) %s", i, jvms()->of_depth(i)->method()->name()->as_utf8());
    }
  }
#endif
  return true;
}

// Helper routine for above
bool LibraryCallKit::is_method_invoke_or_aux_frame(JVMState* jvms) {
3889 3890
  ciMethod* method = jvms->method();

D
duke 已提交
3891
  // Is this the Method.invoke method itself?
3892
  if (method->intrinsic_id() == vmIntrinsics::_invoke)
D
duke 已提交
3893 3894 3895
    return true;

  // Is this a helper, defined somewhere underneath MethodAccessorImpl.
3896
  ciKlass* k = method->holder();
D
duke 已提交
3897 3898 3899 3900 3901 3902 3903 3904 3905
  if (k->is_instance_klass()) {
    ciInstanceKlass* ik = k->as_instance_klass();
    for (; ik != NULL; ik = ik->super()) {
      if (ik->name() == ciSymbol::sun_reflect_MethodAccessorImpl() &&
          ik == env()->find_system_klass(ik->name())) {
        return true;
      }
    }
  }
3906 3907 3908 3909
  else if (method->is_method_handle_adapter()) {
    // This is an internal adapter frame from the MethodHandleCompiler -- skip it
    return true;
  }
D
duke 已提交
3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991

  return false;
}

static int value_field_offset = -1;  // offset of the "value" field of AtomicLongCSImpl.  This is needed by
                                     // inline_native_AtomicLong_attemptUpdate() but it has no way of
                                     // computing it since there is no lookup field by name function in the
                                     // CI interface.  This is computed and set by inline_native_AtomicLong_get().
                                     // Using a static variable here is safe even if we have multiple compilation
                                     // threads because the offset is constant.  At worst the same offset will be
                                     // computed and  stored multiple

bool LibraryCallKit::inline_native_AtomicLong_get() {
  // Restore the stack and pop off the argument
  _sp+=1;
  Node *obj = pop();

  // get the offset of the "value" field. Since the CI interfaces
  // does not provide a way to look up a field by name, we scan the bytecodes
  // to get the field index.  We expect the first 2 instructions of the method
  // to be:
  //    0 aload_0
  //    1 getfield "value"
  ciMethod* method = callee();
  if (value_field_offset == -1)
  {
    ciField* value_field;
    ciBytecodeStream iter(method);
    Bytecodes::Code bc = iter.next();

    if ((bc != Bytecodes::_aload_0) &&
              ((bc != Bytecodes::_aload) || (iter.get_index() != 0)))
      return false;
    bc = iter.next();
    if (bc != Bytecodes::_getfield)
      return false;
    bool ignore;
    value_field = iter.get_field(ignore);
    value_field_offset = value_field->offset_in_bytes();
  }

  // Null check without removing any arguments.
  _sp++;
  obj = do_null_check(obj, T_OBJECT);
  _sp--;
  // Check for locking null object
  if (stopped()) return true;

  Node *adr = basic_plus_adr(obj, obj, value_field_offset);
  const TypePtr *adr_type = _gvn.type(adr)->is_ptr();
  int alias_idx = C->get_alias_index(adr_type);

  Node *result = _gvn.transform(new (C, 3) LoadLLockedNode(control(), memory(alias_idx), adr));

  push_pair(result);

  return true;
}

bool LibraryCallKit::inline_native_AtomicLong_attemptUpdate() {
  // Restore the stack and pop off the arguments
  _sp+=5;
  Node *newVal = pop_pair();
  Node *oldVal = pop_pair();
  Node *obj = pop();

  // we need the offset of the "value" field which was computed when
  // inlining the get() method.  Give up if we don't have it.
  if (value_field_offset == -1)
    return false;

  // Null check without removing any arguments.
  _sp+=5;
  obj = do_null_check(obj, T_OBJECT);
  _sp-=5;
  // Check for locking null object
  if (stopped()) return true;

  Node *adr = basic_plus_adr(obj, obj, value_field_offset);
  const TypePtr *adr_type = _gvn.type(adr)->is_ptr();
  int alias_idx = C->get_alias_index(adr_type);

3992 3993
  Node *cas = _gvn.transform(new (C, 5) StoreLConditionalNode(control(), memory(alias_idx), adr, newVal, oldVal));
  Node *store_proj = _gvn.transform( new (C, 1) SCMemProjNode(cas));
D
duke 已提交
3994
  set_memory(store_proj, alias_idx);
3995
  Node *bol = _gvn.transform( new (C, 2) BoolNode( cas, BoolTest::eq ) );
D
duke 已提交
3996

3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017
  Node *result;
  // CMove node is not used to be able fold a possible check code
  // after attemptUpdate() call. This code could be transformed
  // into CMove node by loop optimizations.
  {
    RegionNode *r = new (C, 3) RegionNode(3);
    result = new (C, 3) PhiNode(r, TypeInt::BOOL);

    Node *iff = create_and_xform_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
    Node *iftrue = opt_iff(r, iff);
    r->init_req(1, iftrue);
    result->init_req(1, intcon(1));
    result->init_req(2, intcon(0));

    set_control(_gvn.transform(r));
    record_for_igvn(r);

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

  push(_gvn.transform(result));
D
duke 已提交
4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189
  return true;
}

bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
  // restore the arguments
  _sp += arg_size();

  switch (id) {
  case vmIntrinsics::_floatToRawIntBits:
    push(_gvn.transform( new (C, 2) MoveF2INode(pop())));
    break;

  case vmIntrinsics::_intBitsToFloat:
    push(_gvn.transform( new (C, 2) MoveI2FNode(pop())));
    break;

  case vmIntrinsics::_doubleToRawLongBits:
    push_pair(_gvn.transform( new (C, 2) MoveD2LNode(pop_pair())));
    break;

  case vmIntrinsics::_longBitsToDouble:
    push_pair(_gvn.transform( new (C, 2) MoveL2DNode(pop_pair())));
    break;

  case vmIntrinsics::_doubleToLongBits: {
    Node* value = pop_pair();

    // two paths (plus control) merge in a wood
    RegionNode *r = new (C, 3) RegionNode(3);
    Node *phi = new (C, 3) PhiNode(r, TypeLong::LONG);

    Node *cmpisnan = _gvn.transform( new (C, 3) CmpDNode(value, value));
    // Build the boolean node
    Node *bolisnan = _gvn.transform( new (C, 2) BoolNode( cmpisnan, BoolTest::ne ) );

    // 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;
    Node *iftrue = _gvn.transform( new (C, 1) IfTrueNode(opt_ifisnan) );

    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
    Node *iffalse = _gvn.transform( new (C, 1) IfFalseNode(opt_ifisnan) );
    set_control(iffalse);

    phi->init_req(2, _gvn.transform( new (C, 2) MoveD2LNode(value)));
    r->init_req(2, iffalse);

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

    Node* result = _gvn.transform(phi);
    assert(result->bottom_type()->isa_long(), "must be");
    push_pair(result);

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

    break;
  }

  case vmIntrinsics::_floatToIntBits: {
    Node* value = pop();

    // two paths (plus control) merge in a wood
    RegionNode *r = new (C, 3) RegionNode(3);
    Node *phi = new (C, 3) PhiNode(r, TypeInt::INT);

    Node *cmpisnan = _gvn.transform( new (C, 3) CmpFNode(value, value));
    // Build the boolean node
    Node *bolisnan = _gvn.transform( new (C, 2) BoolNode( cmpisnan, BoolTest::ne ) );

    // 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;
    Node *iftrue = _gvn.transform( new (C, 1) IfTrueNode(opt_ifisnan) );

    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
    Node *iffalse = _gvn.transform( new (C, 1) IfFalseNode(opt_ifisnan) );
    set_control(iffalse);

    phi->init_req(2, _gvn.transform( new (C, 2) MoveF2INode(value)));
    r->init_req(2, iffalse);

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

    Node* result = _gvn.transform(phi);
    assert(result->bottom_type()->isa_int(), "must be");
    push(result);

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

    break;
  }

  default:
    ShouldNotReachHere();
  }

  return true;
}

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

//----------------------inline_unsafe_copyMemory-------------------------
bool LibraryCallKit::inline_unsafe_copyMemory() {
  if (callee()->is_static())  return false;  // caller must have the capability!
  int nargs = 1 + 5 + 3;  // 5 args:  (src: ptr,off, dst: ptr,off, size)
  assert(signature()->size() == nargs-1, "copy has 5 arguments");
  null_check_receiver(callee());  // check then ignore argument(0)
  if (stopped())  return true;

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

  Node* src_ptr = argument(1);
  Node* src_off = ConvL2X(argument(2));
  assert(argument(3)->is_top(), "2nd half of long");
  Node* dst_ptr = argument(4);
  Node* dst_off = ConvL2X(argument(5));
  assert(argument(6)->is_top(), "2nd half of long");
  Node* size    = ConvL2X(argument(7));
  assert(argument(8)->is_top(), "2nd half of long");

  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;
}

4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207
//------------------------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(), "");

  if (ReduceBulkZeroing) {
    // We will be completely responsible for initializing this object -
    // mark Initialize node as complete.
    AllocateNode* alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
    // The object was just allocated - there should be no any stores!
    guarantee(alloc != NULL && alloc->maybe_set_complete(&_gvn), "");
  }

  // Copy the fastest available way.
  // TODO: generate fields copies for small objects instead.
  Node* src  = obj;
4208
  Node* dest = alloc_obj;
4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240
  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
  // 12 - 64-bit VM, compressed oops
  // 16 - 64-bit VM, normal oops
  if (base_off % BytesPerLong != 0) {
    assert(UseCompressedOops, "");
    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;
  countx = _gvn.transform( new (C, 3) SubXNode(countx, MakeConX(base_off)) );
  countx = _gvn.transform( new (C, 3) URShiftXNode(countx, intcon(LogBytesPerLong) ));

  const TypePtr* raw_adr_type = TypeRawPtr::BOTTOM;
  bool disjoint_bases = true;
  generate_unchecked_arraycopy(raw_adr_type, T_LONG, disjoint_bases,
4241 4242
                               src, NULL, dest, NULL, countx,
                               /*dest_uninitialized*/true);
4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254

  // 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),
4255
                 alloc_obj,
4256 4257 4258 4259 4260 4261 4262
                 no_particular_field,
                 raw_adr_idx,
                 no_particular_value,
                 T_OBJECT,
                 false);
  }

4263 4264
  // Do not let reads from the cloned object float above the arraycopy.
  insert_mem_bar(Op_MemBarCPUOrder);
4265
}
D
duke 已提交
4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284

//------------------------inline_native_clone----------------------------
// 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) {
  int nargs = 1;
4285
  PhiNode* result_val;
D
duke 已提交
4286

4287 4288 4289 4290 4291
  //set the original stack and the reexecute bit for the interpreter to reexecute
  //the bytecode that invokes Object.clone if deoptimization happens
  { PreserveReexecuteState preexecs(this);
    jvms()->set_should_reexecute(true);

4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326
    //null_check_receiver will adjust _sp (push and pop)
    Node* obj = null_check_receiver(callee());
    if (stopped())  return true;

    _sp += nargs;

    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
    };
    RegionNode* result_reg = new(C, PATH_LIMIT) RegionNode(PATH_LIMIT);
    result_val             = new(C, PATH_LIMIT) PhiNode(result_reg,
                                                        TypeInstPtr::NOTNULL);
    PhiNode*    result_i_o = new(C, PATH_LIMIT) PhiNode(result_reg, Type::ABIO);
    PhiNode*    result_mem = new(C, PATH_LIMIT) PhiNode(result_reg, Type::MEMORY,
                                                        TypePtr::BOTTOM);
    record_for_igvn(result_reg);

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

4327 4328 4329 4330 4331 4332 4333
    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;
4334
      Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size);
4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356

      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.
4357 4358 4359 4360 4361 4362
      // (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.)
4363 4364 4365 4366 4367 4368 4369 4370 4371

      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
4372 4373
      }
    }
4374

4375 4376 4377 4378
    // 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.
    RegionNode* slow_region = new (C, 1) RegionNode(1);
    record_for_igvn(slow_region);
4379
    if (!stopped()) {
4380 4381 4382 4383 4384 4385 4386 4387 4388 4389
      // 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);
      }
4390

4391 4392 4393 4394 4395 4396 4397 4398 4399
      // 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);
4400
    }
D
duke 已提交
4401

4402 4403 4404 4405
    if (!stopped()) {
      // It's an instance, and it passed the slow-path tests.
      PreserveJVMState pjvms(this);
      Node* obj_size  = NULL;
4406
      Node* alloc_obj = new_instance(obj_klass, NULL, &obj_size);
4407 4408 4409 4410 4411 4412 4413 4414

      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
4415 4416
    }

4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428
    // 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
4429

4430 4431 4432 4433 4434
    // Return the combined state.
    set_control(    _gvn.transform(result_reg) );
    set_i_o(        _gvn.transform(result_i_o) );
    set_all_memory( _gvn.transform(result_mem) );
  } //original reexecute and sp are set back here
D
duke 已提交
4435

4436
  push(_gvn.transform(result_val));
D
duke 已提交
4437 4438 4439 4440 4441 4442 4443 4444 4445

  return true;
}

//------------------------------basictype2arraycopy----------------------------
address LibraryCallKit::basictype2arraycopy(BasicType t,
                                            Node* src_offset,
                                            Node* dest_offset,
                                            bool disjoint_bases,
4446 4447
                                            const char* &name,
                                            bool dest_uninitialized) {
D
duke 已提交
4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463
  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();
4464
    int element_size = type2aelembytes(t);
D
duke 已提交
4465 4466 4467 4468 4469 4470 4471 4472
    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;
  }

4473
  return StubRoutines::select_arraycopy_function(t, aligned, disjoint, name, dest_uninitialized);
D
duke 已提交
4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505
}


//------------------------------inline_arraycopy-----------------------
bool LibraryCallKit::inline_arraycopy() {
  // Restore the stack and pop off the arguments.
  int nargs = 5;  // 2 oops, 3 ints, no size_t or long
  assert(callee()->signature()->size() == nargs, "copy has 5 arguments");

  Node *src         = argument(0);
  Node *src_offset  = argument(1);
  Node *dest        = argument(2);
  Node *dest_offset = argument(3);
  Node *length      = argument(4);

  // 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.
  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();
  if (top_src  == NULL || top_src->klass()  == NULL ||
      top_dest == NULL || top_dest->klass() == NULL) {
    // 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,
4506
                       src, src_offset, dest, dest_offset, length);
D
duke 已提交
4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528

    // 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,
4529 4530
                            src, src_offset, dest, dest_offset, length,
                            /*dest_uninitialized*/false);
D
duke 已提交
4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586
    return true;
  }

  //---------------------------------------------------------------------------
  // 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

  RegionNode* slow_region = new (C, 1) RegionNode(1);
  record_for_igvn(slow_region);

  // (3) operands must not be null
  // We currently perform our null checks with the do_null_check routine.
  // 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.
  _sp += nargs;
  src  = do_null_check(src,  T_ARRAY);
  dest = do_null_check(dest, T_ARRAY);
  _sp -= nargs;

  // (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,
4587
                     false, false, slow_region);
D
duke 已提交
4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642

  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) {
    slow_region = new(C,1) RegionNode(1);
    record_for_igvn(slow_region);
  }

  Node* original_dest      = dest;
  AllocateArrayNode* alloc = NULL;  // used for zeroing, if needed
4643
  bool  dest_uninitialized = false;
D
duke 已提交
4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655

  // 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
      && !_gvn.eqv_uncast(src, dest)
      && ((alloc = tightly_coupled_allocation(dest, slow_region))
          != NULL)
4656
      && _gvn.find_int_con(alloc->in(AllocateNode::ALength), 1) > 0
D
duke 已提交
4657 4658 4659 4660
      && alloc->maybe_set_complete(&_gvn)) {
    // "You break it, you buy it."
    InitializeNode* init = alloc->initialization();
    assert(init->is_complete(), "we just did this");
4661
    init->set_complete_with_arraycopy();
4662
    assert(dest->is_CheckCastPP(), "sanity");
D
duke 已提交
4663 4664 4665 4666
    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.
4667 4668 4669
    // 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
4670 4671 4672 4673
  } else {
    // No zeroing elimination here.
    alloc             = NULL;
    //original_dest   = dest;
4674
    //dest_uninitialized = false;
D
duke 已提交
4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705
  }

  // 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
  };
  RegionNode* result_region = new(C, PATH_LIMIT) RegionNode(PATH_LIMIT);
  PhiNode*    result_i_o    = new(C, PATH_LIMIT) PhiNode(result_region, Type::ABIO);
  PhiNode*    result_memory = new(C, PATH_LIMIT) PhiNode(result_region, Type::MEMORY, adr_type);
  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) {
4706
    assert(!dest_uninitialized, "");
D
duke 已提交
4707 4708
    Node* cv = generate_generic_arraycopy(adr_type,
                                          src, src_offset, dest, dest_offset,
4709
                                          copy_length, dest_uninitialized);
D
duke 已提交
4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727
    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);
    }

4728
    // copy_length is 0.
4729
    if (!stopped() && dest_uninitialized) {
D
duke 已提交
4730 4731 4732
      Node* dest_length = alloc->in(AllocateNode::ALength);
      if (_gvn.eqv_uncast(copy_length, dest_length)
          || _gvn.find_int_con(dest_length, 1) <= 0) {
4733
        // There is no zeroing to do. No need for a secondary raw memory barrier.
D
duke 已提交
4734 4735 4736 4737 4738
      } 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));
4739 4740 4741 4742 4743 4744 4745
        // 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
4746 4747 4748 4749 4750 4751 4752 4753 4754
      }
    }

    // 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));
  }

4755
  if (!stopped() && dest_uninitialized) {
D
duke 已提交
4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790
    // 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);
    Node* dest_tail   = _gvn.transform( new(C,3) AddINode(dest_offset,
                                                          copy_length) );

    // 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;
    if (!stopped() && !_gvn.eqv_uncast(dest_tail, dest_length)) {
      Node* cmp_lt   = _gvn.transform( new(C,3) CmpINode(dest_tail, dest_length) );
      Node* bol_lt   = _gvn.transform( new(C,2) BoolNode(cmp_lt, BoolTest::lt) );
      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,
4791
                                         dest_size, dest_uninitialized);
D
duke 已提交
4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 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
        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.
        Node* done_ctl = new(C,3) RegionNode(3);
        Node* done_mem = new(C,3) PhiNode(done_ctl, Type::MEMORY, adr_type);
        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));
        set_control( _gvn.transform(done_ctl) );
        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.)

    // Get the klassOop for both src and dest
    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.)
      int ek_offset = objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc);
      Node* p1 = basic_plus_adr(dest_klass, ek_offset);
4862
      Node* n1 = LoadKlassNode::make(_gvn, immutable_memory(), p1, TypeRawPtr::BOTTOM);
D
duke 已提交
4863 4864 4865 4866
      Node* dest_elem_klass = _gvn.transform(n1);
      Node* cv = generate_checkcast_arraycopy(adr_type,
                                              dest_elem_klass,
                                              src, src_offset, dest, dest_offset,
4867
                                              ConvI2X(copy_length), dest_uninitialized);
D
duke 已提交
4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878
      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.
4879
      copy_type = LP64_ONLY(UseCompressedOops ? T_INT : T_LONG) NOT_LP64(T_INT);
4880
      assert(type2aelembytes(basic_elem_type) == type2aelembytes(copy_type),
D
duke 已提交
4881 4882 4883 4884 4885 4886 4887 4888 4889
             "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,
4890
                                 ConvI2X(copy_length), dest_uninitialized);
D
duke 已提交
4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928

    // 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);
  debug_only(slow_region = (RegionNode*)badAddress);

  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.
    Node* cmp = _gvn.transform( new(C, 3) CmpINode(checked_value, intcon(0)) );
    Node* bol = _gvn.transform( new(C, 2) BoolNode(cmp, BoolTest::eq) );
    IfNode* iff = create_and_map_if(control(), bol, PROB_MAX, COUNT_UNKNOWN);

    // If it is 0, we are done, so transfer to the end.
    Node* checks_done = _gvn.transform( new(C, 1) IfTrueNode(iff) );
    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.
    set_control( _gvn.transform( new(C, 1) IfFalseNode(iff) ));
    RegionNode* slow_reg2 = new(C, 3) RegionNode(3);
    PhiNode*    slow_i_o2 = new(C, 3) PhiNode(slow_reg2, Type::ABIO);
    PhiNode*    slow_mem2 = new(C, 3) PhiNode(slow_reg2, Type::MEMORY, adr_type);
    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());
4929 4930
    slow_i_o2  ->init_req(2, checked_i_o);
    slow_mem2  ->init_req(2, checked_mem);
D
duke 已提交
4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968

    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.
      Node* checked_offset = _gvn.transform( new(C, 3) XorINode(checked_value, intcon(-1)) );
      Node* slow_offset    = new(C, 3) PhiNode(slow_reg2, TypeInt::INT);
      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.
      Node* src_off_plus  = _gvn.transform( new(C, 3) AddINode(src_offset,  slow_offset) );
      Node* dest_off_plus = _gvn.transform( new(C, 3) AddINode(dest_offset, slow_offset) );
      Node* length_minus  = _gvn.transform( new(C, 3) SubINode(copy_length, slow_offset) );

      // 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);

4969
    if (dest_uninitialized) {
D
duke 已提交
4970 4971 4972 4973 4974 4975 4976
      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,
4977
                            copy_length, /*dest_uninitialized*/false);
D
duke 已提交
4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995

    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.
  set_control( _gvn.transform(result_region) );
  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
4996
  // array copies accurately to allow value numbering of field loads around
D
duke 已提交
4997 4998 4999 5000 5001 5002 5003 5004
  // 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.
5005 5006 5007
  //
  // Do not let reads from the cloned object float above the arraycopy.
  if (InsertMemBarAfterArraycopy || alloc != NULL)
D
duke 已提交
5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067
    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
5068
            (obs->as_Call()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
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 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126
          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:
5127
  int scale = exact_log2(type2aelembytes(basic_elem_type));
D
duke 已提交
5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181
  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)
      end = _gvn.transform( new(C,3) LShiftXNode(end, intcon(scale) ));
    end_base += end_round;
    end = _gvn.transform( new(C,3) AddXNode(end, MakeConX(end_base)) );
    end = _gvn.transform( new(C,3) AndXNode(end, MakeConX(~end_round)) );
    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)
      start = _gvn.transform( new(C,3) LShiftXNode( start, intcon(scale) ));
    start = _gvn.transform( new(C,3) AddXNode(start, MakeConX(abase)) );
    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.
5182 5183 5184 5185 5186 5187 5188 5189 5190
      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:
        start = _gvn.transform( new(C,3) AddXNode(start, MakeConX(bump_bit)) );
        assert((abase & clear_low) == 0, "array base must be int-aligned");
      }
D
duke 已提交
5191 5192
      // Round bumped 'start' down to jlong boundary in body of array.
      start = _gvn.transform( new(C,3) AndXNode(start, MakeConX(~to_clear)) );
5193 5194 5195 5196 5197 5198 5199
      if (bump_bit != 0) {
        // Store a zero to the immediately preceding jint:
        Node* x1 = _gvn.transform( new(C,3) AddXNode(start, MakeConX(-bump_bit)) );
        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
5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220
    }
    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,
5221
                                         Node* dest_size, bool dest_uninitialized) {
D
duke 已提交
5222
  // See if there is an advantage from block transfer.
5223
  int scale = exact_log2(type2aelembytes(basic_elem_type));
D
duke 已提交
5224 5225 5226 5227 5228 5229
  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);

5230 5231 5232
  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
5233 5234 5235
    // At present, we can only understand constants.
    return false;

5236 5237 5238
  intptr_t src_off  = abase + (src_off_con  << scale);
  intptr_t dest_off = abase + (dest_off_con << scale);

D
duke 已提交
5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253 5254 5255 5256 5257 5258 5259 5260 5261 5262 5263 5264 5265 5266
  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;
  countx = _gvn.transform( new (C, 3) SubXNode(countx, MakeConX(dest_off)) );
  countx = _gvn.transform( new (C, 3) URShiftXNode(countx, intcon(LogBytesPerLong)) );

  bool disjoint_bases = true;   // since alloc != NULL
  generate_unchecked_arraycopy(adr_type, T_LONG, disjoint_bases,
5267
                               sptr, NULL, dptr, NULL, countx, dest_uninitialized);
D
duke 已提交
5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279

  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,
5280 5281
                                        Node* copy_length, bool dest_uninitialized) {
  assert(!dest_uninitialized, "Invariant");
D
duke 已提交
5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294 5295 5296 5297 5298
  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,
5299
                                             Node* copy_length, bool dest_uninitialized) {
D
duke 已提交
5300 5301
  if (stopped())  return NULL;

5302
  address copyfunc_addr = StubRoutines::checkcast_arraycopy(dest_uninitialized);
D
duke 已提交
5303 5304 5305 5306 5307 5308 5309 5310 5311 5312
  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.
  int sco_offset = Klass::super_check_offset_offset_in_bytes() + sizeof(oopDesc);
  Node* p3 = basic_plus_adr(dest_elem_klass, sco_offset);
5313
  Node* n3 = new(C, 3) LoadINode(NULL, memory(p3), p3, _gvn.type(p3)->is_ptr());
5314
  Node* check_offset = ConvI2X(_gvn.transform(n3));
D
duke 已提交
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
  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);

  return _gvn.transform(new (C, 1) ProjNode(call, TypeFunc::Parms));
}


// 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,
5340 5341
                                           Node* copy_length, bool dest_uninitialized) {
  assert(!dest_uninitialized, "Invariant");
D
duke 已提交
5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362
  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);

  return _gvn.transform(new (C, 1) ProjNode(call, TypeFunc::Parms));
}

// 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,
5363
                                             Node* copy_length, bool dest_uninitialized) {
D
duke 已提交
5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377
  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,
5378
                          disjoint_bases, copyfunc_name, dest_uninitialized);
D
duke 已提交
5379 5380 5381 5382 5383 5384 5385

  // 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);
}
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 5413 5414 5415 5416 5417 5418

//----------------------------inline_reference_get----------------------------

bool LibraryCallKit::inline_reference_get() {
  const int nargs = 1; // self

  guarantee(java_lang_ref_Reference::referent_offset > 0,
            "should have already been set");

  int referent_offset = java_lang_ref_Reference::referent_offset;

  // Restore the stack and pop off the argument
  _sp += nargs;
  Node *reference_obj = pop();

  // Null check on self without removing any arguments.
  _sp += nargs;
  reference_obj = do_null_check(reference_obj, T_OBJECT);
  _sp -= nargs;;

  if (stopped()) return true;

  Node *adr = basic_plus_adr(reference_obj, reference_obj, referent_offset);

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

  Node* no_ctrl = NULL;
  Node *result = make_load(no_ctrl, adr, object_type, T_OBJECT);

  // Use the pre-barrier to record the value in the referent field
  pre_barrier(false /* do_load */,
              control(),
5419
              NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,
5420 5421 5422 5423 5424 5425 5426
              result /* pre_val */,
              T_OBJECT);

  push(result);
  return true;
}