1 /*
   2  * Copyright (c) 2005, 2011, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "ci/bcEscapeAnalyzer.hpp"
  27 #include "libadt/vectset.hpp"
  28 #include "memory/allocation.hpp"
  29 #include "opto/c2compiler.hpp"
  30 #include "opto/callnode.hpp"
  31 #include "opto/cfgnode.hpp"
  32 #include "opto/compile.hpp"
  33 #include "opto/escape.hpp"
  34 #include "opto/phaseX.hpp"
  35 #include "opto/rootnode.hpp"
  36 
  37 void PointsToNode::add_edge(uint targIdx, PointsToNode::EdgeType et) {
  38   uint v = (targIdx << EdgeShift) + ((uint) et);
  39   if (_edges == NULL) {
  40      Arena *a = Compile::current()->comp_arena();
  41     _edges = new(a) GrowableArray<uint>(a, INITIAL_EDGE_COUNT, 0, 0);
  42   }
  43   _edges->append_if_missing(v);
  44 }
  45 
  46 void PointsToNode::remove_edge(uint targIdx, PointsToNode::EdgeType et) {
  47   uint v = (targIdx << EdgeShift) + ((uint) et);
  48 
  49   _edges->remove(v);
  50 }
  51 
  52 #ifndef PRODUCT
  53 static const char *node_type_names[] = {
  54   "UnknownType",
  55   "JavaObject",
  56   "LocalVar",
  57   "Field"
  58 };
  59 
  60 static const char *esc_names[] = {
  61   "UnknownEscape",
  62   "NoEscape",
  63   "ArgEscape",
  64   "GlobalEscape"
  65 };
  66 
  67 static const char *edge_type_suffix[] = {
  68  "?", // UnknownEdge
  69  "P", // PointsToEdge
  70  "D", // DeferredEdge
  71  "F"  // FieldEdge
  72 };
  73 
  74 void PointsToNode::dump(bool print_state) const {
  75   NodeType nt = node_type();
  76   tty->print("%s ", node_type_names[(int) nt]);
  77   if (print_state) {
  78     EscapeState es = escape_state();
  79     tty->print("%s %s ", esc_names[(int) es], _scalar_replaceable ? "":"NSR");
  80   }
  81   tty->print("[[");
  82   for (uint i = 0; i < edge_count(); i++) {
  83     tty->print(" %d%s", edge_target(i), edge_type_suffix[(int) edge_type(i)]);
  84   }
  85   tty->print("]]  ");
  86   if (_node == NULL)
  87     tty->print_cr("<null>");
  88   else
  89     _node->dump();
  90 }
  91 #endif
  92 
  93 ConnectionGraph::ConnectionGraph(Compile * C, PhaseIterGVN *igvn) :
  94   _nodes(C->comp_arena(), C->unique(), C->unique(), PointsToNode()),
  95   _processed(C->comp_arena()),
  96   pt_ptset(C->comp_arena()),
  97   pt_visited(C->comp_arena()),
  98   pt_worklist(C->comp_arena(), 4, 0, 0),
  99   _collecting(true),
 100   _progress(false),
 101   _compile(C),
 102   _igvn(igvn),
 103   _node_map(C->comp_arena()) {
 104 
 105   _phantom_object = C->top()->_idx,
 106   add_node(C->top(), PointsToNode::JavaObject, PointsToNode::GlobalEscape,true);
 107 
 108   // Add ConP(#NULL) and ConN(#NULL) nodes.
 109   Node* oop_null = igvn->zerocon(T_OBJECT);
 110   _oop_null = oop_null->_idx;
 111   assert(_oop_null < nodes_size(), "should be created already");
 112   add_node(oop_null, PointsToNode::JavaObject, PointsToNode::NoEscape, true);
 113 
 114   if (UseCompressedOops) {
 115     Node* noop_null = igvn->zerocon(T_NARROWOOP);
 116     _noop_null = noop_null->_idx;
 117     assert(_noop_null < nodes_size(), "should be created already");
 118     add_node(noop_null, PointsToNode::JavaObject, PointsToNode::NoEscape, true);
 119   } else {
 120     _noop_null = _oop_null; // Should be initialized
 121   }
 122   _pcmp_neq = NULL; // Should be initialized
 123   _pcmp_eq  = NULL;
 124 }
 125 
 126 void ConnectionGraph::add_pointsto_edge(uint from_i, uint to_i) {
 127   PointsToNode *f = ptnode_adr(from_i);
 128   PointsToNode *t = ptnode_adr(to_i);
 129 
 130   assert(f->node_type() != PointsToNode::UnknownType && t->node_type() != PointsToNode::UnknownType, "node types must be set");
 131   assert(f->node_type() == PointsToNode::LocalVar || f->node_type() == PointsToNode::Field, "invalid source of PointsTo edge");
 132   assert(t->node_type() == PointsToNode::JavaObject, "invalid destination of PointsTo edge");
 133   add_edge(f, to_i, PointsToNode::PointsToEdge);
 134 }
 135 
 136 void ConnectionGraph::add_deferred_edge(uint from_i, uint to_i) {
 137   PointsToNode *f = ptnode_adr(from_i);
 138   PointsToNode *t = ptnode_adr(to_i);
 139 
 140   assert(f->node_type() != PointsToNode::UnknownType && t->node_type() != PointsToNode::UnknownType, "node types must be set");
 141   assert(f->node_type() == PointsToNode::LocalVar || f->node_type() == PointsToNode::Field, "invalid source of Deferred edge");
 142   assert(t->node_type() == PointsToNode::LocalVar || t->node_type() == PointsToNode::Field, "invalid destination of Deferred edge");
 143   // don't add a self-referential edge, this can occur during removal of
 144   // deferred edges
 145   if (from_i != to_i)
 146     add_edge(f, to_i, PointsToNode::DeferredEdge);
 147 }
 148 
 149 int ConnectionGraph::address_offset(Node* adr, PhaseTransform *phase) {
 150   const Type *adr_type = phase->type(adr);
 151   if (adr->is_AddP() && adr_type->isa_oopptr() == NULL &&
 152       adr->in(AddPNode::Address)->is_Proj() &&
 153       adr->in(AddPNode::Address)->in(0)->is_Allocate()) {
 154     // We are computing a raw address for a store captured by an Initialize
 155     // compute an appropriate address type. AddP cases #3 and #5 (see below).
 156     int offs = (int)phase->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
 157     assert(offs != Type::OffsetBot ||
 158            adr->in(AddPNode::Address)->in(0)->is_AllocateArray(),
 159            "offset must be a constant or it is initialization of array");
 160     return offs;
 161   }
 162   const TypePtr *t_ptr = adr_type->isa_ptr();
 163   assert(t_ptr != NULL, "must be a pointer type");
 164   return t_ptr->offset();
 165 }
 166 
 167 void ConnectionGraph::add_field_edge(uint from_i, uint to_i, int offset) {
 168   PointsToNode *f = ptnode_adr(from_i);
 169   PointsToNode *t = ptnode_adr(to_i);
 170 
 171   assert(f->node_type() != PointsToNode::UnknownType && t->node_type() != PointsToNode::UnknownType, "node types must be set");
 172   assert(f->node_type() == PointsToNode::JavaObject, "invalid destination of Field edge");
 173   assert(t->node_type() == PointsToNode::Field, "invalid destination of Field edge");
 174   assert (t->offset() == -1 || t->offset() == offset, "conflicting field offsets");
 175   t->set_offset(offset);
 176 
 177   add_edge(f, to_i, PointsToNode::FieldEdge);
 178 }
 179 
 180 void ConnectionGraph::set_escape_state(uint ni, PointsToNode::EscapeState es) {
 181   // Don't change non-escaping state of NULL pointer.
 182   if (ni == _noop_null || ni == _oop_null)
 183     return;
 184   PointsToNode *npt = ptnode_adr(ni);
 185   PointsToNode::EscapeState old_es = npt->escape_state();
 186   if (es > old_es)
 187     npt->set_escape_state(es);
 188 }
 189 
 190 void ConnectionGraph::add_node(Node *n, PointsToNode::NodeType nt,
 191                                PointsToNode::EscapeState es, bool done) {
 192   PointsToNode* ptadr = ptnode_adr(n->_idx);
 193   ptadr->_node = n;
 194   ptadr->set_node_type(nt);
 195 
 196   // inline set_escape_state(idx, es);
 197   PointsToNode::EscapeState old_es = ptadr->escape_state();
 198   if (es > old_es)
 199     ptadr->set_escape_state(es);
 200 
 201   if (done)
 202     _processed.set(n->_idx);
 203 }
 204 
 205 PointsToNode::EscapeState ConnectionGraph::escape_state(Node *n) {
 206   uint idx = n->_idx;
 207   PointsToNode::EscapeState es;
 208 
 209   // If we are still collecting or there were no non-escaping allocations
 210   // we don't know the answer yet
 211   if (_collecting)
 212     return PointsToNode::UnknownEscape;
 213 
 214   // if the node was created after the escape computation, return
 215   // UnknownEscape
 216   if (idx >= nodes_size())
 217     return PointsToNode::UnknownEscape;
 218 
 219   es = ptnode_adr(idx)->escape_state();
 220 
 221   // if we have already computed a value, return it
 222   if (es != PointsToNode::UnknownEscape &&
 223       ptnode_adr(idx)->node_type() == PointsToNode::JavaObject)
 224     return es;
 225 
 226   // PointsTo() calls n->uncast() which can return a new ideal node.
 227   if (n->uncast()->_idx >= nodes_size())
 228     return PointsToNode::UnknownEscape;
 229 
 230   PointsToNode::EscapeState orig_es = es;
 231 
 232   // compute max escape state of anything this node could point to
 233   for(VectorSetI i(PointsTo(n)); i.test() && es != PointsToNode::GlobalEscape; ++i) {
 234     uint pt = i.elem;
 235     PointsToNode::EscapeState pes = ptnode_adr(pt)->escape_state();
 236     if (pes > es)
 237       es = pes;
 238   }
 239   if (orig_es != es) {
 240     // cache the computed escape state
 241     assert(es > orig_es, "should have computed an escape state");
 242     set_escape_state(idx, es);
 243   } // orig_es could be PointsToNode::UnknownEscape
 244   return es;
 245 }
 246 
 247 VectorSet* ConnectionGraph::PointsTo(Node * n) {
 248   pt_ptset.Reset();
 249   pt_visited.Reset();
 250   pt_worklist.clear();
 251 
 252 #ifdef ASSERT
 253   Node *orig_n = n;
 254 #endif
 255 
 256   n = n->uncast();
 257   PointsToNode* npt = ptnode_adr(n->_idx);
 258 
 259   // If we have a JavaObject, return just that object
 260   if (npt->node_type() == PointsToNode::JavaObject) {
 261     pt_ptset.set(n->_idx);
 262     return &pt_ptset;
 263   }
 264 #ifdef ASSERT
 265   if (npt->_node == NULL) {
 266     if (orig_n != n)
 267       orig_n->dump();
 268     n->dump();
 269     assert(npt->_node != NULL, "unregistered node");
 270   }
 271 #endif
 272   pt_worklist.push(n->_idx);
 273   while(pt_worklist.length() > 0) {
 274     int ni = pt_worklist.pop();
 275     if (pt_visited.test_set(ni))
 276       continue;
 277 
 278     PointsToNode* pn = ptnode_adr(ni);
 279     // ensure that all inputs of a Phi have been processed
 280     assert(!_collecting || !pn->_node->is_Phi() || _processed.test(ni),"");
 281 
 282     int edges_processed = 0;
 283     uint e_cnt = pn->edge_count();
 284     for (uint e = 0; e < e_cnt; e++) {
 285       uint etgt = pn->edge_target(e);
 286       PointsToNode::EdgeType et = pn->edge_type(e);
 287       if (et == PointsToNode::PointsToEdge) {
 288         pt_ptset.set(etgt);
 289         edges_processed++;
 290       } else if (et == PointsToNode::DeferredEdge) {
 291         pt_worklist.push(etgt);
 292         edges_processed++;
 293       } else {
 294         assert(false,"neither PointsToEdge or DeferredEdge");
 295       }
 296     }
 297     if (edges_processed == 0) {
 298       // no deferred or pointsto edges found.  Assume the value was set
 299       // outside this method.  Add the phantom object to the pointsto set.
 300       pt_ptset.set(_phantom_object);
 301     }
 302   }
 303   return &pt_ptset;
 304 }
 305 
 306 void ConnectionGraph::remove_deferred(uint ni, GrowableArray<uint>* deferred_edges, VectorSet* visited) {
 307   // This method is most expensive during ConnectionGraph construction.
 308   // Reuse vectorSet and an additional growable array for deferred edges.
 309   deferred_edges->clear();
 310   visited->Reset();
 311 
 312   visited->set(ni);
 313   PointsToNode *ptn = ptnode_adr(ni);
 314   if (ptn->edge_count() == 0) {
 315     // No deferred or pointsto edges found.  Assume the value was set
 316     // outside this method.  Add edge to phantom object.
 317     add_pointsto_edge(ni, _phantom_object);
 318   }
 319 
 320   // Mark current edges as visited and move deferred edges to separate array.
 321   for (uint i = 0; i < ptn->edge_count(); ) {
 322     uint t = ptn->edge_target(i);
 323 #ifdef ASSERT
 324     assert(!visited->test_set(t), "expecting no duplications");
 325 #else
 326     visited->set(t);
 327 #endif
 328     if (ptn->edge_type(i) == PointsToNode::DeferredEdge) {
 329       ptn->remove_edge(t, PointsToNode::DeferredEdge);
 330       deferred_edges->append(t);
 331     } else {
 332       i++;
 333     }
 334   }
 335   for (int next = 0; next < deferred_edges->length(); ++next) {
 336     uint t = deferred_edges->at(next);
 337     PointsToNode *ptt = ptnode_adr(t);
 338     uint e_cnt = ptt->edge_count();
 339     if (e_cnt == 0) {
 340       // No deferred or pointsto edges found.  Assume the value was set
 341       // outside this method.  Add edge to phantom object.
 342       add_pointsto_edge(t, _phantom_object);
 343       add_pointsto_edge(ni, _phantom_object);
 344     }
 345     for (uint e = 0; e < e_cnt; e++) {
 346       uint etgt = ptt->edge_target(e);
 347       if (visited->test_set(etgt))
 348         continue;
 349 
 350       PointsToNode::EdgeType et = ptt->edge_type(e);
 351       if (et == PointsToNode::PointsToEdge) {
 352         add_pointsto_edge(ni, etgt);
 353         if(etgt == _phantom_object) {
 354           // Special case - field set outside (globally escaping).
 355           set_escape_state(ni, PointsToNode::GlobalEscape);
 356         }
 357       } else if (et == PointsToNode::DeferredEdge) {
 358         deferred_edges->append(etgt);
 359       } else {
 360         assert(false,"invalid connection graph");
 361       }
 362     }
 363   }
 364 }
 365 
 366 
 367 //  Add an edge to node given by "to_i" from any field of adr_i whose offset
 368 //  matches "offset"  A deferred edge is added if to_i is a LocalVar, and
 369 //  a pointsto edge is added if it is a JavaObject
 370 
 371 void ConnectionGraph::add_edge_from_fields(uint adr_i, uint to_i, int offs) {
 372   PointsToNode* an = ptnode_adr(adr_i);
 373   PointsToNode* to = ptnode_adr(to_i);
 374   bool deferred = (to->node_type() == PointsToNode::LocalVar);
 375 
 376   for (uint fe = 0; fe < an->edge_count(); fe++) {
 377     assert(an->edge_type(fe) == PointsToNode::FieldEdge, "expecting a field edge");
 378     int fi = an->edge_target(fe);
 379     PointsToNode* pf = ptnode_adr(fi);
 380     int po = pf->offset();
 381     if (po == offs || po == Type::OffsetBot || offs == Type::OffsetBot) {
 382       if (deferred)
 383         add_deferred_edge(fi, to_i);
 384       else
 385         add_pointsto_edge(fi, to_i);
 386     }
 387   }
 388 }
 389 
 390 // Add a deferred  edge from node given by "from_i" to any field of adr_i
 391 // whose offset matches "offset".
 392 void ConnectionGraph::add_deferred_edge_to_fields(uint from_i, uint adr_i, int offs) {
 393   PointsToNode* an = ptnode_adr(adr_i);
 394   bool is_alloc = an->_node->is_Allocate();
 395   for (uint fe = 0; fe < an->edge_count(); fe++) {
 396     assert(an->edge_type(fe) == PointsToNode::FieldEdge, "expecting a field edge");
 397     int fi = an->edge_target(fe);
 398     PointsToNode* pf = ptnode_adr(fi);
 399     int offset = pf->offset();
 400     if (!is_alloc) {
 401       // Assume the field was set outside this method if it is not Allocation
 402       add_pointsto_edge(fi, _phantom_object);
 403     }
 404     if (offset == offs || offset == Type::OffsetBot || offs == Type::OffsetBot) {
 405       add_deferred_edge(from_i, fi);
 406     }
 407   }
 408   // Some fields references (AddP) may still be missing
 409   // until Connection Graph construction is complete.
 410   // For example, loads from RAW pointers with offset 0
 411   // which don't have AddP.
 412   // A reference to phantom_object will be added if
 413   // a field reference is still missing after completing
 414   // Connection Graph (see remove_deferred()).
 415 }
 416 
 417 // Helper functions
 418 
 419 static Node* get_addp_base(Node *addp) {
 420   assert(addp->is_AddP(), "must be AddP");
 421   //
 422   // AddP cases for Base and Address inputs:
 423   // case #1. Direct object's field reference:
 424   //     Allocate
 425   //       |
 426   //     Proj #5 ( oop result )
 427   //       |
 428   //     CheckCastPP (cast to instance type)
 429   //      | |
 430   //     AddP  ( base == address )
 431   //
 432   // case #2. Indirect object's field reference:
 433   //      Phi
 434   //       |
 435   //     CastPP (cast to instance type)
 436   //      | |
 437   //     AddP  ( base == address )
 438   //
 439   // case #3. Raw object's field reference for Initialize node:
 440   //      Allocate
 441   //        |
 442   //      Proj #5 ( oop result )
 443   //  top   |
 444   //     \  |
 445   //     AddP  ( base == top )
 446   //
 447   // case #4. Array's element reference:
 448   //   {CheckCastPP | CastPP}
 449   //     |  | |
 450   //     |  AddP ( array's element offset )
 451   //     |  |
 452   //     AddP ( array's offset )
 453   //
 454   // case #5. Raw object's field reference for arraycopy stub call:
 455   //          The inline_native_clone() case when the arraycopy stub is called
 456   //          after the allocation before Initialize and CheckCastPP nodes.
 457   //      Allocate
 458   //        |
 459   //      Proj #5 ( oop result )
 460   //       | |
 461   //       AddP  ( base == address )
 462   //
 463   // case #6. Constant Pool, ThreadLocal, CastX2P or
 464   //          Raw object's field reference:
 465   //      {ConP, ThreadLocal, CastX2P, raw Load}
 466   //  top   |
 467   //     \  |
 468   //     AddP  ( base == top )
 469   //
 470   // case #7. Klass's field reference.
 471   //      LoadKlass
 472   //       | |
 473   //       AddP  ( base == address )
 474   //
 475   // case #8. narrow Klass's field reference.
 476   //      LoadNKlass
 477   //       |
 478   //      DecodeN
 479   //       | |
 480   //       AddP  ( base == address )
 481   //
 482   Node *base = addp->in(AddPNode::Base)->uncast();
 483   if (base->is_top()) { // The AddP case #3 and #6.
 484     base = addp->in(AddPNode::Address)->uncast();
 485     while (base->is_AddP()) {
 486       // Case #6 (unsafe access) may have several chained AddP nodes.
 487       assert(base->in(AddPNode::Base)->is_top(), "expected unsafe access address only");
 488       base = base->in(AddPNode::Address)->uncast();
 489     }
 490     assert(base->Opcode() == Op_ConP || base->Opcode() == Op_ThreadLocal ||
 491            base->Opcode() == Op_CastX2P || base->is_DecodeN() ||
 492            (base->is_Mem() && base->bottom_type() == TypeRawPtr::NOTNULL) ||
 493            (base->is_Proj() && base->in(0)->is_Allocate()), "sanity");
 494   }
 495   return base;
 496 }
 497 
 498 static Node* find_second_addp(Node* addp, Node* n) {
 499   assert(addp->is_AddP() && addp->outcnt() > 0, "Don't process dead nodes");
 500 
 501   Node* addp2 = addp->raw_out(0);
 502   if (addp->outcnt() == 1 && addp2->is_AddP() &&
 503       addp2->in(AddPNode::Base) == n &&
 504       addp2->in(AddPNode::Address) == addp) {
 505 
 506     assert(addp->in(AddPNode::Base) == n, "expecting the same base");
 507     //
 508     // Find array's offset to push it on worklist first and
 509     // as result process an array's element offset first (pushed second)
 510     // to avoid CastPP for the array's offset.
 511     // Otherwise the inserted CastPP (LocalVar) will point to what
 512     // the AddP (Field) points to. Which would be wrong since
 513     // the algorithm expects the CastPP has the same point as
 514     // as AddP's base CheckCastPP (LocalVar).
 515     //
 516     //    ArrayAllocation
 517     //     |
 518     //    CheckCastPP
 519     //     |
 520     //    memProj (from ArrayAllocation CheckCastPP)
 521     //     |  ||
 522     //     |  ||   Int (element index)
 523     //     |  ||    |   ConI (log(element size))
 524     //     |  ||    |   /
 525     //     |  ||   LShift
 526     //     |  ||  /
 527     //     |  AddP (array's element offset)
 528     //     |  |
 529     //     |  | ConI (array's offset: #12(32-bits) or #24(64-bits))
 530     //     | / /
 531     //     AddP (array's offset)
 532     //      |
 533     //     Load/Store (memory operation on array's element)
 534     //
 535     return addp2;
 536   }
 537   return NULL;
 538 }
 539 
 540 //
 541 // Adjust the type and inputs of an AddP which computes the
 542 // address of a field of an instance
 543 //
 544 bool ConnectionGraph::split_AddP(Node *addp, Node *base,  PhaseGVN  *igvn) {
 545   const TypeOopPtr *base_t = igvn->type(base)->isa_oopptr();
 546   assert(base_t != NULL && base_t->is_known_instance(), "expecting instance oopptr");
 547   const TypeOopPtr *t = igvn->type(addp)->isa_oopptr();
 548   if (t == NULL) {
 549     // We are computing a raw address for a store captured by an Initialize
 550     // compute an appropriate address type (cases #3 and #5).
 551     assert(igvn->type(addp) == TypeRawPtr::NOTNULL, "must be raw pointer");
 552     assert(addp->in(AddPNode::Address)->is_Proj(), "base of raw address must be result projection from allocation");
 553     intptr_t offs = (int)igvn->find_intptr_t_con(addp->in(AddPNode::Offset), Type::OffsetBot);
 554     assert(offs != Type::OffsetBot, "offset must be a constant");
 555     t = base_t->add_offset(offs)->is_oopptr();
 556   }
 557   int inst_id =  base_t->instance_id();
 558   assert(!t->is_known_instance() || t->instance_id() == inst_id,
 559                              "old type must be non-instance or match new type");
 560 
 561   // The type 't' could be subclass of 'base_t'.
 562   // As result t->offset() could be large then base_t's size and it will
 563   // cause the failure in add_offset() with narrow oops since TypeOopPtr()
 564   // constructor verifies correctness of the offset.
 565   //
 566   // It could happened on subclass's branch (from the type profiling
 567   // inlining) which was not eliminated during parsing since the exactness
 568   // of the allocation type was not propagated to the subclass type check.
 569   //
 570   // Or the type 't' could be not related to 'base_t' at all.
 571   // It could happened when CHA type is different from MDO type on a dead path
 572   // (for example, from instanceof check) which is not collapsed during parsing.
 573   //
 574   // Do nothing for such AddP node and don't process its users since
 575   // this code branch will go away.
 576   //
 577   if (!t->is_known_instance() &&
 578       !base_t->klass()->is_subtype_of(t->klass())) {
 579      return false; // bail out
 580   }
 581 
 582   const TypeOopPtr *tinst = base_t->add_offset(t->offset())->is_oopptr();
 583   // Do NOT remove the next line: ensure a new alias index is allocated
 584   // for the instance type. Note: C++ will not remove it since the call
 585   // has side effect.
 586   int alias_idx = _compile->get_alias_index(tinst);
 587   igvn->set_type(addp, tinst);
 588   // record the allocation in the node map
 589   assert(ptnode_adr(addp->_idx)->_node != NULL, "should be registered");
 590   set_map(addp->_idx, get_map(base->_idx));
 591 
 592   // Set addp's Base and Address to 'base'.
 593   Node *abase = addp->in(AddPNode::Base);
 594   Node *adr   = addp->in(AddPNode::Address);
 595   if (adr->is_Proj() && adr->in(0)->is_Allocate() &&
 596       adr->in(0)->_idx == (uint)inst_id) {
 597     // Skip AddP cases #3 and #5.
 598   } else {
 599     assert(!abase->is_top(), "sanity"); // AddP case #3
 600     if (abase != base) {
 601       igvn->hash_delete(addp);
 602       addp->set_req(AddPNode::Base, base);
 603       if (abase == adr) {
 604         addp->set_req(AddPNode::Address, base);
 605       } else {
 606         // AddP case #4 (adr is array's element offset AddP node)
 607 #ifdef ASSERT
 608         const TypeOopPtr *atype = igvn->type(adr)->isa_oopptr();
 609         assert(adr->is_AddP() && atype != NULL &&
 610                atype->instance_id() == inst_id, "array's element offset should be processed first");
 611 #endif
 612       }
 613       igvn->hash_insert(addp);
 614     }
 615   }
 616   // Put on IGVN worklist since at least addp's type was changed above.
 617   record_for_optimizer(addp);
 618   return true;
 619 }
 620 
 621 //
 622 // Create a new version of orig_phi if necessary. Returns either the newly
 623 // created phi or an existing phi.  Sets create_new to indicate whether a new
 624 // phi was created.  Cache the last newly created phi in the node map.
 625 //
 626 PhiNode *ConnectionGraph::create_split_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *>  &orig_phi_worklist, PhaseGVN  *igvn, bool &new_created) {
 627   Compile *C = _compile;
 628   new_created = false;
 629   int phi_alias_idx = C->get_alias_index(orig_phi->adr_type());
 630   // nothing to do if orig_phi is bottom memory or matches alias_idx
 631   if (phi_alias_idx == alias_idx) {
 632     return orig_phi;
 633   }
 634   // Have we recently created a Phi for this alias index?
 635   PhiNode *result = get_map_phi(orig_phi->_idx);
 636   if (result != NULL && C->get_alias_index(result->adr_type()) == alias_idx) {
 637     return result;
 638   }
 639   // Previous check may fail when the same wide memory Phi was split into Phis
 640   // for different memory slices. Search all Phis for this region.
 641   if (result != NULL) {
 642     Node* region = orig_phi->in(0);
 643     for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
 644       Node* phi = region->fast_out(i);
 645       if (phi->is_Phi() &&
 646           C->get_alias_index(phi->as_Phi()->adr_type()) == alias_idx) {
 647         assert(phi->_idx >= nodes_size(), "only new Phi per instance memory slice");
 648         return phi->as_Phi();
 649       }
 650     }
 651   }
 652   if ((int)C->unique() + 2*NodeLimitFudgeFactor > MaxNodeLimit) {
 653     if (C->do_escape_analysis() == true && !C->failing()) {
 654       // Retry compilation without escape analysis.
 655       // If this is the first failure, the sentinel string will "stick"
 656       // to the Compile object, and the C2Compiler will see it and retry.
 657       C->record_failure(C2Compiler::retry_no_escape_analysis());
 658     }
 659     return NULL;
 660   }
 661   orig_phi_worklist.append_if_missing(orig_phi);
 662   const TypePtr *atype = C->get_adr_type(alias_idx);
 663   result = PhiNode::make(orig_phi->in(0), NULL, Type::MEMORY, atype);
 664   C->copy_node_notes_to(result, orig_phi);
 665   igvn->set_type(result, result->bottom_type());
 666   record_for_optimizer(result);
 667 
 668   debug_only(Node* pn = ptnode_adr(orig_phi->_idx)->_node;)
 669   assert(pn == NULL || pn == orig_phi, "wrong node");
 670   set_map(orig_phi->_idx, result);
 671   ptnode_adr(orig_phi->_idx)->_node = orig_phi;
 672 
 673   new_created = true;
 674   return result;
 675 }
 676 
 677 //
 678 // Return a new version of Memory Phi "orig_phi" with the inputs having the
 679 // specified alias index.
 680 //
 681 PhiNode *ConnectionGraph::split_memory_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *>  &orig_phi_worklist, PhaseGVN  *igvn) {
 682 
 683   assert(alias_idx != Compile::AliasIdxBot, "can't split out bottom memory");
 684   Compile *C = _compile;
 685   bool new_phi_created;
 686   PhiNode *result = create_split_phi(orig_phi, alias_idx, orig_phi_worklist, igvn, new_phi_created);
 687   if (!new_phi_created) {
 688     return result;
 689   }
 690 
 691   GrowableArray<PhiNode *>  phi_list;
 692   GrowableArray<uint>  cur_input;
 693 
 694   PhiNode *phi = orig_phi;
 695   uint idx = 1;
 696   bool finished = false;
 697   while(!finished) {
 698     while (idx < phi->req()) {
 699       Node *mem = find_inst_mem(phi->in(idx), alias_idx, orig_phi_worklist, igvn);
 700       if (mem != NULL && mem->is_Phi()) {
 701         PhiNode *newphi = create_split_phi(mem->as_Phi(), alias_idx, orig_phi_worklist, igvn, new_phi_created);
 702         if (new_phi_created) {
 703           // found an phi for which we created a new split, push current one on worklist and begin
 704           // processing new one
 705           phi_list.push(phi);
 706           cur_input.push(idx);
 707           phi = mem->as_Phi();
 708           result = newphi;
 709           idx = 1;
 710           continue;
 711         } else {
 712           mem = newphi;
 713         }
 714       }
 715       if (C->failing()) {
 716         return NULL;
 717       }
 718       result->set_req(idx++, mem);
 719     }
 720 #ifdef ASSERT
 721     // verify that the new Phi has an input for each input of the original
 722     assert( phi->req() == result->req(), "must have same number of inputs.");
 723     assert( result->in(0) != NULL && result->in(0) == phi->in(0), "regions must match");
 724 #endif
 725     // Check if all new phi's inputs have specified alias index.
 726     // Otherwise use old phi.
 727     for (uint i = 1; i < phi->req(); i++) {
 728       Node* in = result->in(i);
 729       assert((phi->in(i) == NULL) == (in == NULL), "inputs must correspond.");
 730     }
 731     // we have finished processing a Phi, see if there are any more to do
 732     finished = (phi_list.length() == 0 );
 733     if (!finished) {
 734       phi = phi_list.pop();
 735       idx = cur_input.pop();
 736       PhiNode *prev_result = get_map_phi(phi->_idx);
 737       prev_result->set_req(idx++, result);
 738       result = prev_result;
 739     }
 740   }
 741   return result;
 742 }
 743 
 744 
 745 //
 746 // The next methods are derived from methods in MemNode.
 747 //
 748 static Node *step_through_mergemem(MergeMemNode *mmem, int alias_idx, const TypeOopPtr *toop) {
 749   Node *mem = mmem;
 750   // TypeOopPtr::NOTNULL+any is an OOP with unknown offset - generally
 751   // means an array I have not precisely typed yet.  Do not do any
 752   // alias stuff with it any time soon.
 753   if( toop->base() != Type::AnyPtr &&
 754       !(toop->klass() != NULL &&
 755         toop->klass()->is_java_lang_Object() &&
 756         toop->offset() == Type::OffsetBot) ) {
 757     mem = mmem->memory_at(alias_idx);
 758     // Update input if it is progress over what we have now
 759   }
 760   return mem;
 761 }
 762 
 763 //
 764 // Move memory users to their memory slices.
 765 //
 766 void ConnectionGraph::move_inst_mem(Node* n, GrowableArray<PhiNode *>  &orig_phis, PhaseGVN *igvn) {
 767   Compile* C = _compile;
 768 
 769   const TypePtr* tp = igvn->type(n->in(MemNode::Address))->isa_ptr();
 770   assert(tp != NULL, "ptr type");
 771   int alias_idx = C->get_alias_index(tp);
 772   int general_idx = C->get_general_index(alias_idx);
 773 
 774   // Move users first
 775   for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
 776     Node* use = n->fast_out(i);
 777     if (use->is_MergeMem()) {
 778       MergeMemNode* mmem = use->as_MergeMem();
 779       assert(n == mmem->memory_at(alias_idx), "should be on instance memory slice");
 780       if (n != mmem->memory_at(general_idx) || alias_idx == general_idx) {
 781         continue; // Nothing to do
 782       }
 783       // Replace previous general reference to mem node.
 784       uint orig_uniq = C->unique();
 785       Node* m = find_inst_mem(n, general_idx, orig_phis, igvn);
 786       assert(orig_uniq == C->unique(), "no new nodes");
 787       mmem->set_memory_at(general_idx, m);
 788       --imax;
 789       --i;
 790     } else if (use->is_MemBar()) {
 791       assert(!use->is_Initialize(), "initializing stores should not be moved");
 792       if (use->req() > MemBarNode::Precedent &&
 793           use->in(MemBarNode::Precedent) == n) {
 794         // Don't move related membars.
 795         record_for_optimizer(use);
 796         continue;
 797       }
 798       tp = use->as_MemBar()->adr_type()->isa_ptr();
 799       if (tp != NULL && C->get_alias_index(tp) == alias_idx ||
 800           alias_idx == general_idx) {
 801         continue; // Nothing to do
 802       }
 803       // Move to general memory slice.
 804       uint orig_uniq = C->unique();
 805       Node* m = find_inst_mem(n, general_idx, orig_phis, igvn);
 806       assert(orig_uniq == C->unique(), "no new nodes");
 807       igvn->hash_delete(use);
 808       imax -= use->replace_edge(n, m);
 809       igvn->hash_insert(use);
 810       record_for_optimizer(use);
 811       --i;
 812 #ifdef ASSERT
 813     } else if (use->is_Mem()) {
 814       if (use->Opcode() == Op_StoreCM && use->in(MemNode::OopStore) == n) {
 815         // Don't move related cardmark.
 816         continue;
 817       }
 818       // Memory nodes should have new memory input.
 819       tp = igvn->type(use->in(MemNode::Address))->isa_ptr();
 820       assert(tp != NULL, "ptr type");
 821       int idx = C->get_alias_index(tp);
 822       assert(get_map(use->_idx) != NULL || idx == alias_idx,
 823              "Following memory nodes should have new memory input or be on the same memory slice");
 824     } else if (use->is_Phi()) {
 825       // Phi nodes should be split and moved already.
 826       tp = use->as_Phi()->adr_type()->isa_ptr();
 827       assert(tp != NULL, "ptr type");
 828       int idx = C->get_alias_index(tp);
 829       assert(idx == alias_idx, "Following Phi nodes should be on the same memory slice");
 830     } else {
 831       use->dump();
 832       assert(false, "should not be here");
 833 #endif
 834     }
 835   }
 836 }
 837 
 838 //
 839 // Search memory chain of "mem" to find a MemNode whose address
 840 // is the specified alias index.
 841 //
 842 Node* ConnectionGraph::find_inst_mem(Node *orig_mem, int alias_idx, GrowableArray<PhiNode *>  &orig_phis, PhaseGVN *phase) {
 843   if (orig_mem == NULL)
 844     return orig_mem;
 845   Compile* C = phase->C;
 846   const TypeOopPtr *toop = C->get_adr_type(alias_idx)->isa_oopptr();
 847   bool is_instance = (toop != NULL) && toop->is_known_instance();
 848   Node *start_mem = C->start()->proj_out(TypeFunc::Memory);
 849   Node *prev = NULL;
 850   Node *result = orig_mem;
 851   while (prev != result) {
 852     prev = result;
 853     if (result == start_mem)
 854       break;  // hit one of our sentinels
 855     if (result->is_Mem()) {
 856       const Type *at = phase->type(result->in(MemNode::Address));
 857       if (at == Type::TOP)
 858         break; // Dead
 859       assert (at->isa_ptr() != NULL, "pointer type required.");
 860       int idx = C->get_alias_index(at->is_ptr());
 861       if (idx == alias_idx)
 862         break; // Found
 863       if (!is_instance && (at->isa_oopptr() == NULL ||
 864                            !at->is_oopptr()->is_known_instance())) {
 865         break; // Do not skip store to general memory slice.
 866       }
 867       result = result->in(MemNode::Memory);
 868     }
 869     if (!is_instance)
 870       continue;  // don't search further for non-instance types
 871     // skip over a call which does not affect this memory slice
 872     if (result->is_Proj() && result->as_Proj()->_con == TypeFunc::Memory) {
 873       Node *proj_in = result->in(0);
 874       if (proj_in->is_Allocate() && proj_in->_idx == (uint)toop->instance_id()) {
 875         break;  // hit one of our sentinels
 876       } else if (proj_in->is_Call()) {
 877         CallNode *call = proj_in->as_Call();
 878         if (!call->may_modify(toop, phase)) {
 879           result = call->in(TypeFunc::Memory);
 880         }
 881       } else if (proj_in->is_Initialize()) {
 882         AllocateNode* alloc = proj_in->as_Initialize()->allocation();
 883         // Stop if this is the initialization for the object instance which
 884         // which contains this memory slice, otherwise skip over it.
 885         if (alloc == NULL || alloc->_idx != (uint)toop->instance_id()) {
 886           result = proj_in->in(TypeFunc::Memory);
 887         }
 888       } else if (proj_in->is_MemBar()) {
 889         result = proj_in->in(TypeFunc::Memory);
 890       }
 891     } else if (result->is_MergeMem()) {
 892       MergeMemNode *mmem = result->as_MergeMem();
 893       result = step_through_mergemem(mmem, alias_idx, toop);
 894       if (result == mmem->base_memory()) {
 895         // Didn't find instance memory, search through general slice recursively.
 896         result = mmem->memory_at(C->get_general_index(alias_idx));
 897         result = find_inst_mem(result, alias_idx, orig_phis, phase);
 898         if (C->failing()) {
 899           return NULL;
 900         }
 901         mmem->set_memory_at(alias_idx, result);
 902       }
 903     } else if (result->is_Phi() &&
 904                C->get_alias_index(result->as_Phi()->adr_type()) != alias_idx) {
 905       Node *un = result->as_Phi()->unique_input(phase);
 906       if (un != NULL) {
 907         orig_phis.append_if_missing(result->as_Phi());
 908         result = un;
 909       } else {
 910         break;
 911       }
 912     } else if (result->is_ClearArray()) {
 913       if (!ClearArrayNode::step_through(&result, (uint)toop->instance_id(), phase)) {
 914         // Can not bypass initialization of the instance
 915         // we are looking for.
 916         break;
 917       }
 918       // Otherwise skip it (the call updated 'result' value).
 919     } else if (result->Opcode() == Op_SCMemProj) {
 920       assert(result->in(0)->is_LoadStore(), "sanity");
 921       const Type *at = phase->type(result->in(0)->in(MemNode::Address));
 922       if (at != Type::TOP) {
 923         assert (at->isa_ptr() != NULL, "pointer type required.");
 924         int idx = C->get_alias_index(at->is_ptr());
 925         assert(idx != alias_idx, "Object is not scalar replaceable if a LoadStore node access its field");
 926         break;
 927       }
 928       result = result->in(0)->in(MemNode::Memory);
 929     }
 930   }
 931   if (result->is_Phi()) {
 932     PhiNode *mphi = result->as_Phi();
 933     assert(mphi->bottom_type() == Type::MEMORY, "memory phi required");
 934     const TypePtr *t = mphi->adr_type();
 935     if (!is_instance) {
 936       // Push all non-instance Phis on the orig_phis worklist to update inputs
 937       // during Phase 4 if needed.
 938       orig_phis.append_if_missing(mphi);
 939     } else if (C->get_alias_index(t) != alias_idx) {
 940       // Create a new Phi with the specified alias index type.
 941       result = split_memory_phi(mphi, alias_idx, orig_phis, phase);
 942     }
 943   }
 944   // the result is either MemNode, PhiNode, InitializeNode.
 945   return result;
 946 }
 947 
 948 //
 949 //  Convert the types of unescaped object to instance types where possible,
 950 //  propagate the new type information through the graph, and update memory
 951 //  edges and MergeMem inputs to reflect the new type.
 952 //
 953 //  We start with allocations (and calls which may be allocations)  on alloc_worklist.
 954 //  The processing is done in 4 phases:
 955 //
 956 //  Phase 1:  Process possible allocations from alloc_worklist.  Create instance
 957 //            types for the CheckCastPP for allocations where possible.
 958 //            Propagate the the new types through users as follows:
 959 //               casts and Phi:  push users on alloc_worklist
 960 //               AddP:  cast Base and Address inputs to the instance type
 961 //                      push any AddP users on alloc_worklist and push any memnode
 962 //                      users onto memnode_worklist.
 963 //  Phase 2:  Process MemNode's from memnode_worklist. compute new address type and
 964 //            search the Memory chain for a store with the appropriate type
 965 //            address type.  If a Phi is found, create a new version with
 966 //            the appropriate memory slices from each of the Phi inputs.
 967 //            For stores, process the users as follows:
 968 //               MemNode:  push on memnode_worklist
 969 //               MergeMem: push on mergemem_worklist
 970 //  Phase 3:  Process MergeMem nodes from mergemem_worklist.  Walk each memory slice
 971 //            moving the first node encountered of each  instance type to the
 972 //            the input corresponding to its alias index.
 973 //            appropriate memory slice.
 974 //  Phase 4:  Update the inputs of non-instance memory Phis and the Memory input of memnodes.
 975 //
 976 // In the following example, the CheckCastPP nodes are the cast of allocation
 977 // results and the allocation of node 29 is unescaped and eligible to be an
 978 // instance type.
 979 //
 980 // We start with:
 981 //
 982 //     7 Parm #memory
 983 //    10  ConI  "12"
 984 //    19  CheckCastPP   "Foo"
 985 //    20  AddP  _ 19 19 10  Foo+12  alias_index=4
 986 //    29  CheckCastPP   "Foo"
 987 //    30  AddP  _ 29 29 10  Foo+12  alias_index=4
 988 //
 989 //    40  StoreP  25   7  20   ... alias_index=4
 990 //    50  StoreP  35  40  30   ... alias_index=4
 991 //    60  StoreP  45  50  20   ... alias_index=4
 992 //    70  LoadP    _  60  30   ... alias_index=4
 993 //    80  Phi     75  50  60   Memory alias_index=4
 994 //    90  LoadP    _  80  30   ... alias_index=4
 995 //   100  LoadP    _  80  20   ... alias_index=4
 996 //
 997 //
 998 // Phase 1 creates an instance type for node 29 assigning it an instance id of 24
 999 // and creating a new alias index for node 30.  This gives:
1000 //
1001 //     7 Parm #memory
1002 //    10  ConI  "12"
1003 //    19  CheckCastPP   "Foo"
1004 //    20  AddP  _ 19 19 10  Foo+12  alias_index=4
1005 //    29  CheckCastPP   "Foo"  iid=24
1006 //    30  AddP  _ 29 29 10  Foo+12  alias_index=6  iid=24
1007 //
1008 //    40  StoreP  25   7  20   ... alias_index=4
1009 //    50  StoreP  35  40  30   ... alias_index=6
1010 //    60  StoreP  45  50  20   ... alias_index=4
1011 //    70  LoadP    _  60  30   ... alias_index=6
1012 //    80  Phi     75  50  60   Memory alias_index=4
1013 //    90  LoadP    _  80  30   ... alias_index=6
1014 //   100  LoadP    _  80  20   ... alias_index=4
1015 //
1016 // In phase 2, new memory inputs are computed for the loads and stores,
1017 // And a new version of the phi is created.  In phase 4, the inputs to
1018 // node 80 are updated and then the memory nodes are updated with the
1019 // values computed in phase 2.  This results in:
1020 //
1021 //     7 Parm #memory
1022 //    10  ConI  "12"
1023 //    19  CheckCastPP   "Foo"
1024 //    20  AddP  _ 19 19 10  Foo+12  alias_index=4
1025 //    29  CheckCastPP   "Foo"  iid=24
1026 //    30  AddP  _ 29 29 10  Foo+12  alias_index=6  iid=24
1027 //
1028 //    40  StoreP  25  7   20   ... alias_index=4
1029 //    50  StoreP  35  7   30   ... alias_index=6
1030 //    60  StoreP  45  40  20   ... alias_index=4
1031 //    70  LoadP    _  50  30   ... alias_index=6
1032 //    80  Phi     75  40  60   Memory alias_index=4
1033 //   120  Phi     75  50  50   Memory alias_index=6
1034 //    90  LoadP    _ 120  30   ... alias_index=6
1035 //   100  LoadP    _  80  20   ... alias_index=4
1036 //
1037 void ConnectionGraph::split_unique_types(GrowableArray<Node *>  &alloc_worklist) {
1038   GrowableArray<Node *>  memnode_worklist;
1039   GrowableArray<PhiNode *>  orig_phis;
1040 
1041   PhaseIterGVN  *igvn = _igvn;
1042   uint new_index_start = (uint) _compile->num_alias_types();
1043   Arena* arena = Thread::current()->resource_area();
1044   VectorSet visited(arena);
1045 
1046 
1047   //  Phase 1:  Process possible allocations from alloc_worklist.
1048   //  Create instance types for the CheckCastPP for allocations where possible.
1049   //
1050   // (Note: don't forget to change the order of the second AddP node on
1051   //  the alloc_worklist if the order of the worklist processing is changed,
1052   //  see the comment in find_second_addp().)
1053   //
1054   while (alloc_worklist.length() != 0) {
1055     Node *n = alloc_worklist.pop();
1056     uint ni = n->_idx;
1057     const TypeOopPtr* tinst = NULL;
1058     if (n->is_Call()) {
1059       CallNode *alloc = n->as_Call();
1060       // copy escape information to call node
1061       PointsToNode* ptn = ptnode_adr(alloc->_idx);
1062       PointsToNode::EscapeState es = escape_state(alloc);
1063       // We have an allocation or call which returns a Java object,
1064       // see if it is unescaped.
1065       if (es != PointsToNode::NoEscape || !ptn->scalar_replaceable())
1066         continue;
1067 
1068       // Find CheckCastPP for the allocate or for the return value of a call
1069       n = alloc->result_cast();
1070       if (n == NULL) {            // No uses except Initialize node
1071         if (alloc->is_Allocate()) {
1072           // Set the scalar_replaceable flag for allocation
1073           // so it could be eliminated if it has no uses.
1074           alloc->as_Allocate()->_is_scalar_replaceable = true;
1075         }
1076         continue;
1077       }
1078       if (!n->is_CheckCastPP()) { // not unique CheckCastPP.
1079         assert(!alloc->is_Allocate(), "allocation should have unique type");
1080         continue;
1081       }
1082 
1083       // The inline code for Object.clone() casts the allocation result to
1084       // java.lang.Object and then to the actual type of the allocated
1085       // object. Detect this case and use the second cast.
1086       // Also detect j.l.reflect.Array.newInstance(jobject, jint) case when
1087       // the allocation result is cast to java.lang.Object and then
1088       // to the actual Array type.
1089       if (alloc->is_Allocate() && n->as_Type()->type() == TypeInstPtr::NOTNULL
1090           && (alloc->is_AllocateArray() ||
1091               igvn->type(alloc->in(AllocateNode::KlassNode)) != TypeKlassPtr::OBJECT)) {
1092         Node *cast2 = NULL;
1093         for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1094           Node *use = n->fast_out(i);
1095           if (use->is_CheckCastPP()) {
1096             cast2 = use;
1097             break;
1098           }
1099         }
1100         if (cast2 != NULL) {
1101           n = cast2;
1102         } else {
1103           // Non-scalar replaceable if the allocation type is unknown statically
1104           // (reflection allocation), the object can't be restored during
1105           // deoptimization without precise type.
1106           continue;
1107         }
1108       }
1109       if (alloc->is_Allocate()) {
1110         // Set the scalar_replaceable flag for allocation
1111         // so it could be eliminated.
1112         alloc->as_Allocate()->_is_scalar_replaceable = true;
1113       }
1114       set_escape_state(n->_idx, es); // CheckCastPP escape state
1115       // in order for an object to be scalar-replaceable, it must be:
1116       //   - a direct allocation (not a call returning an object)
1117       //   - non-escaping
1118       //   - eligible to be a unique type
1119       //   - not determined to be ineligible by escape analysis
1120       assert(ptnode_adr(alloc->_idx)->_node != NULL &&
1121              ptnode_adr(n->_idx)->_node != NULL, "should be registered");
1122       set_map(alloc->_idx, n);
1123       set_map(n->_idx, alloc);
1124       const TypeOopPtr *t = igvn->type(n)->isa_oopptr();
1125       if (t == NULL)
1126         continue;  // not a TypeOopPtr
1127       tinst = t->cast_to_exactness(true)->is_oopptr()->cast_to_instance_id(ni);
1128       igvn->hash_delete(n);
1129       igvn->set_type(n,  tinst);
1130       n->raise_bottom_type(tinst);
1131       igvn->hash_insert(n);
1132       record_for_optimizer(n);
1133       if (alloc->is_Allocate() && (t->isa_instptr() || t->isa_aryptr())) {
1134 
1135         // First, put on the worklist all Field edges from Connection Graph
1136         // which is more accurate then putting immediate users from Ideal Graph.
1137         for (uint e = 0; e < ptn->edge_count(); e++) {
1138           Node *use = ptnode_adr(ptn->edge_target(e))->_node;
1139           assert(ptn->edge_type(e) == PointsToNode::FieldEdge && use->is_AddP(),
1140                  "only AddP nodes are Field edges in CG");
1141           if (use->outcnt() > 0) { // Don't process dead nodes
1142             Node* addp2 = find_second_addp(use, use->in(AddPNode::Base));
1143             if (addp2 != NULL) {
1144               assert(alloc->is_AllocateArray(),"array allocation was expected");
1145               alloc_worklist.append_if_missing(addp2);
1146             }
1147             alloc_worklist.append_if_missing(use);
1148           }
1149         }
1150 
1151         // An allocation may have an Initialize which has raw stores. Scan
1152         // the users of the raw allocation result and push AddP users
1153         // on alloc_worklist.
1154         Node *raw_result = alloc->proj_out(TypeFunc::Parms);
1155         assert (raw_result != NULL, "must have an allocation result");
1156         for (DUIterator_Fast imax, i = raw_result->fast_outs(imax); i < imax; i++) {
1157           Node *use = raw_result->fast_out(i);
1158           if (use->is_AddP() && use->outcnt() > 0) { // Don't process dead nodes
1159             Node* addp2 = find_second_addp(use, raw_result);
1160             if (addp2 != NULL) {
1161               assert(alloc->is_AllocateArray(),"array allocation was expected");
1162               alloc_worklist.append_if_missing(addp2);
1163             }
1164             alloc_worklist.append_if_missing(use);
1165           } else if (use->is_MemBar()) {
1166             memnode_worklist.append_if_missing(use);
1167           }
1168         }
1169       }
1170     } else if (n->is_AddP()) {
1171       VectorSet* ptset = PointsTo(get_addp_base(n));
1172       assert(ptset->Size() == 1, "AddP address is unique");
1173       uint elem = ptset->getelem(); // Allocation node's index
1174       if (elem == _phantom_object) {
1175         assert(false, "escaped allocation");
1176         continue; // Assume the value was set outside this method.
1177       }
1178       Node *base = get_map(elem);  // CheckCastPP node
1179       if (!split_AddP(n, base, igvn)) continue; // wrong type from dead path
1180       tinst = igvn->type(base)->isa_oopptr();
1181     } else if (n->is_Phi() ||
1182                n->is_CheckCastPP() ||
1183                n->is_EncodeP() ||
1184                n->is_DecodeN() ||
1185                (n->is_ConstraintCast() && n->Opcode() == Op_CastPP)) {
1186       if (visited.test_set(n->_idx)) {
1187         assert(n->is_Phi(), "loops only through Phi's");
1188         continue;  // already processed
1189       }
1190       VectorSet* ptset = PointsTo(n);
1191       if (ptset->Size() == 1) {
1192         uint elem = ptset->getelem(); // Allocation node's index
1193         if (elem == _phantom_object) {
1194           assert(false, "escaped allocation");
1195           continue; // Assume the value was set outside this method.
1196         }
1197         Node *val = get_map(elem);   // CheckCastPP node
1198         TypeNode *tn = n->as_Type();
1199         tinst = igvn->type(val)->isa_oopptr();
1200         assert(tinst != NULL && tinst->is_known_instance() &&
1201                (uint)tinst->instance_id() == elem , "instance type expected.");
1202 
1203         const Type *tn_type = igvn->type(tn);
1204         const TypeOopPtr *tn_t;
1205         if (tn_type->isa_narrowoop()) {
1206           tn_t = tn_type->make_ptr()->isa_oopptr();
1207         } else {
1208           tn_t = tn_type->isa_oopptr();
1209         }
1210 
1211         if (tn_t != NULL && tinst->klass()->is_subtype_of(tn_t->klass())) {
1212           if (tn_type->isa_narrowoop()) {
1213             tn_type = tinst->make_narrowoop();
1214           } else {
1215             tn_type = tinst;
1216           }
1217           igvn->hash_delete(tn);
1218           igvn->set_type(tn, tn_type);
1219           tn->set_type(tn_type);
1220           igvn->hash_insert(tn);
1221           record_for_optimizer(n);
1222         } else {
1223           assert(tn_type == TypePtr::NULL_PTR ||
1224                  tn_t != NULL && !tinst->klass()->is_subtype_of(tn_t->klass()),
1225                  "unexpected type");
1226           continue; // Skip dead path with different type
1227         }
1228       }
1229     } else {
1230       debug_only(n->dump();)
1231       assert(false, "EA: unexpected node");
1232       continue;
1233     }
1234     // push allocation's users on appropriate worklist
1235     for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1236       Node *use = n->fast_out(i);
1237       if(use->is_Mem() && use->in(MemNode::Address) == n) {
1238         // Load/store to instance's field
1239         memnode_worklist.append_if_missing(use);
1240       } else if (use->is_MemBar()) {
1241         memnode_worklist.append_if_missing(use);
1242       } else if (use->is_AddP() && use->outcnt() > 0) { // No dead nodes
1243         Node* addp2 = find_second_addp(use, n);
1244         if (addp2 != NULL) {
1245           alloc_worklist.append_if_missing(addp2);
1246         }
1247         alloc_worklist.append_if_missing(use);
1248       } else if (use->is_Phi() ||
1249                  use->is_CheckCastPP() ||
1250                  use->is_EncodeP() ||
1251                  use->is_DecodeN() ||
1252                  (use->is_ConstraintCast() && use->Opcode() == Op_CastPP)) {
1253         alloc_worklist.append_if_missing(use);
1254 #ifdef ASSERT
1255       } else if (use->is_Mem()) {
1256         assert(use->in(MemNode::Address) != n, "EA: missing allocation reference path");
1257       } else if (use->is_MergeMem()) {
1258         assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist");
1259       } else if (use->is_SafePoint()) {
1260         // Look for MergeMem nodes for calls which reference unique allocation
1261         // (through CheckCastPP nodes) even for debug info.
1262         Node* m = use->in(TypeFunc::Memory);
1263         if (m->is_MergeMem()) {
1264           assert(_mergemem_worklist.contains(m->as_MergeMem()), "EA: missing MergeMem node in the worklist");
1265         }
1266       } else {
1267         uint op = use->Opcode();
1268         if (!(op == Op_CmpP || op == Op_Conv2B ||
1269               op == Op_CastP2X || op == Op_StoreCM ||
1270               op == Op_FastLock || op == Op_AryEq || op == Op_StrComp ||
1271               op == Op_StrEquals || op == Op_StrIndexOf)) {
1272           n->dump();
1273           use->dump();
1274           assert(false, "EA: missing allocation reference path");
1275         }
1276 #endif
1277       }
1278     }
1279 
1280   }
1281   // New alias types were created in split_AddP().
1282   uint new_index_end = (uint) _compile->num_alias_types();
1283 
1284   //  Phase 2:  Process MemNode's from memnode_worklist. compute new address type and
1285   //            compute new values for Memory inputs  (the Memory inputs are not
1286   //            actually updated until phase 4.)
1287   if (memnode_worklist.length() == 0)
1288     return;  // nothing to do
1289 
1290   while (memnode_worklist.length() != 0) {
1291     Node *n = memnode_worklist.pop();
1292     if (visited.test_set(n->_idx))
1293       continue;
1294     if (n->is_Phi() || n->is_ClearArray()) {
1295       // we don't need to do anything, but the users must be pushed
1296     } else if (n->is_MemBar()) { // Initialize, MemBar nodes
1297       // we don't need to do anything, but the users must be pushed
1298       n = n->as_MemBar()->proj_out(TypeFunc::Memory);
1299       if (n == NULL)
1300         continue;
1301     } else {
1302       assert(n->is_Mem(), "memory node required.");
1303       Node *addr = n->in(MemNode::Address);
1304       const Type *addr_t = igvn->type(addr);
1305       if (addr_t == Type::TOP)
1306         continue;
1307       assert (addr_t->isa_ptr() != NULL, "pointer type required.");
1308       int alias_idx = _compile->get_alias_index(addr_t->is_ptr());
1309       assert ((uint)alias_idx < new_index_end, "wrong alias index");
1310       Node *mem = find_inst_mem(n->in(MemNode::Memory), alias_idx, orig_phis, igvn);
1311       if (_compile->failing()) {
1312         return;
1313       }
1314       if (mem != n->in(MemNode::Memory)) {
1315         // We delay the memory edge update since we need old one in
1316         // MergeMem code below when instances memory slices are separated.
1317         debug_only(Node* pn = ptnode_adr(n->_idx)->_node;)
1318         assert(pn == NULL || pn == n, "wrong node");
1319         set_map(n->_idx, mem);
1320         ptnode_adr(n->_idx)->_node = n;
1321       }
1322       if (n->is_Load()) {
1323         continue;  // don't push users
1324       } else if (n->is_LoadStore()) {
1325         // get the memory projection
1326         for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1327           Node *use = n->fast_out(i);
1328           if (use->Opcode() == Op_SCMemProj) {
1329             n = use;
1330             break;
1331           }
1332         }
1333         assert(n->Opcode() == Op_SCMemProj, "memory projection required");
1334       }
1335     }
1336     // push user on appropriate worklist
1337     for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1338       Node *use = n->fast_out(i);
1339       if (use->is_Phi() || use->is_ClearArray()) {
1340         memnode_worklist.append_if_missing(use);
1341       } else if(use->is_Mem() && use->in(MemNode::Memory) == n) {
1342         if (use->Opcode() == Op_StoreCM) // Ignore cardmark stores
1343           continue;
1344         memnode_worklist.append_if_missing(use);
1345       } else if (use->is_MemBar()) {
1346         memnode_worklist.append_if_missing(use);
1347 #ifdef ASSERT
1348       } else if(use->is_Mem()) {
1349         assert(use->in(MemNode::Memory) != n, "EA: missing memory path");
1350       } else if (use->is_MergeMem()) {
1351         assert(_mergemem_worklist.contains(use->as_MergeMem()), "EA: missing MergeMem node in the worklist");
1352       } else {
1353         uint op = use->Opcode();
1354         if (!(op == Op_StoreCM ||
1355               (op == Op_CallLeaf && use->as_CallLeaf()->_name != NULL &&
1356                strcmp(use->as_CallLeaf()->_name, "g1_wb_pre") == 0) ||
1357               op == Op_AryEq || op == Op_StrComp ||
1358               op == Op_StrEquals || op == Op_StrIndexOf)) {
1359           n->dump();
1360           use->dump();
1361           assert(false, "EA: missing memory path");
1362         }
1363 #endif
1364       }
1365     }
1366   }
1367 
1368   //  Phase 3:  Process MergeMem nodes from mergemem_worklist.
1369   //            Walk each memory slice moving the first node encountered of each
1370   //            instance type to the the input corresponding to its alias index.
1371   uint length = _mergemem_worklist.length();
1372   for( uint next = 0; next < length; ++next ) {
1373     MergeMemNode* nmm = _mergemem_worklist.at(next);
1374     assert(!visited.test_set(nmm->_idx), "should not be visited before");
1375     // Note: we don't want to use MergeMemStream here because we only want to
1376     // scan inputs which exist at the start, not ones we add during processing.
1377     // Note 2: MergeMem may already contains instance memory slices added
1378     // during find_inst_mem() call when memory nodes were processed above.
1379     igvn->hash_delete(nmm);
1380     uint nslices = nmm->req();
1381     for (uint i = Compile::AliasIdxRaw+1; i < nslices; i++) {
1382       Node* mem = nmm->in(i);
1383       Node* cur = NULL;
1384       if (mem == NULL || mem->is_top())
1385         continue;
1386       // First, update mergemem by moving memory nodes to corresponding slices
1387       // if their type became more precise since this mergemem was created.
1388       while (mem->is_Mem()) {
1389         const Type *at = igvn->type(mem->in(MemNode::Address));
1390         if (at != Type::TOP) {
1391           assert (at->isa_ptr() != NULL, "pointer type required.");
1392           uint idx = (uint)_compile->get_alias_index(at->is_ptr());
1393           if (idx == i) {
1394             if (cur == NULL)
1395               cur = mem;
1396           } else {
1397             if (idx >= nmm->req() || nmm->is_empty_memory(nmm->in(idx))) {
1398               nmm->set_memory_at(idx, mem);
1399             }
1400           }
1401         }
1402         mem = mem->in(MemNode::Memory);
1403       }
1404       nmm->set_memory_at(i, (cur != NULL) ? cur : mem);
1405       // Find any instance of the current type if we haven't encountered
1406       // already a memory slice of the instance along the memory chain.
1407       for (uint ni = new_index_start; ni < new_index_end; ni++) {
1408         if((uint)_compile->get_general_index(ni) == i) {
1409           Node *m = (ni >= nmm->req()) ? nmm->empty_memory() : nmm->in(ni);
1410           if (nmm->is_empty_memory(m)) {
1411             Node* result = find_inst_mem(mem, ni, orig_phis, igvn);
1412             if (_compile->failing()) {
1413               return;
1414             }
1415             nmm->set_memory_at(ni, result);
1416           }
1417         }
1418       }
1419     }
1420     // Find the rest of instances values
1421     for (uint ni = new_index_start; ni < new_index_end; ni++) {
1422       const TypeOopPtr *tinst = _compile->get_adr_type(ni)->isa_oopptr();
1423       Node* result = step_through_mergemem(nmm, ni, tinst);
1424       if (result == nmm->base_memory()) {
1425         // Didn't find instance memory, search through general slice recursively.
1426         result = nmm->memory_at(_compile->get_general_index(ni));
1427         result = find_inst_mem(result, ni, orig_phis, igvn);
1428         if (_compile->failing()) {
1429           return;
1430         }
1431         nmm->set_memory_at(ni, result);
1432       }
1433     }
1434     igvn->hash_insert(nmm);
1435     record_for_optimizer(nmm);
1436   }
1437 
1438   //  Phase 4:  Update the inputs of non-instance memory Phis and
1439   //            the Memory input of memnodes
1440   // First update the inputs of any non-instance Phi's from
1441   // which we split out an instance Phi.  Note we don't have
1442   // to recursively process Phi's encounted on the input memory
1443   // chains as is done in split_memory_phi() since they  will
1444   // also be processed here.
1445   for (int j = 0; j < orig_phis.length(); j++) {
1446     PhiNode *phi = orig_phis.at(j);
1447     int alias_idx = _compile->get_alias_index(phi->adr_type());
1448     igvn->hash_delete(phi);
1449     for (uint i = 1; i < phi->req(); i++) {
1450       Node *mem = phi->in(i);
1451       Node *new_mem = find_inst_mem(mem, alias_idx, orig_phis, igvn);
1452       if (_compile->failing()) {
1453         return;
1454       }
1455       if (mem != new_mem) {
1456         phi->set_req(i, new_mem);
1457       }
1458     }
1459     igvn->hash_insert(phi);
1460     record_for_optimizer(phi);
1461   }
1462 
1463   // Update the memory inputs of MemNodes with the value we computed
1464   // in Phase 2 and move stores memory users to corresponding memory slices.
1465 
1466   // Disable memory split verification code until the fix for 6984348.
1467   // Currently it produces false negative results since it does not cover all cases.
1468 #if 0 // ifdef ASSERT
1469   visited.Reset();
1470   Node_Stack old_mems(arena, _compile->unique() >> 2);
1471 #endif
1472   for (uint i = 0; i < nodes_size(); i++) {
1473     Node *nmem = get_map(i);
1474     if (nmem != NULL) {
1475       Node *n = ptnode_adr(i)->_node;
1476       assert(n != NULL, "sanity");
1477       if (n->is_Mem()) {
1478 #if 0 // ifdef ASSERT
1479         Node* old_mem = n->in(MemNode::Memory);
1480         if (!visited.test_set(old_mem->_idx)) {
1481           old_mems.push(old_mem, old_mem->outcnt());
1482         }
1483 #endif
1484         assert(n->in(MemNode::Memory) != nmem, "sanity");
1485         if (!n->is_Load()) {
1486           // Move memory users of a store first.
1487           move_inst_mem(n, orig_phis, igvn);
1488         }
1489         // Now update memory input
1490         igvn->hash_delete(n);
1491         n->set_req(MemNode::Memory, nmem);
1492         igvn->hash_insert(n);
1493         record_for_optimizer(n);
1494       } else {
1495         assert(n->is_Allocate() || n->is_CheckCastPP() ||
1496                n->is_AddP() || n->is_Phi(), "unknown node used for set_map()");
1497       }
1498     }
1499   }
1500 #if 0 // ifdef ASSERT
1501   // Verify that memory was split correctly
1502   while (old_mems.is_nonempty()) {
1503     Node* old_mem = old_mems.node();
1504     uint  old_cnt = old_mems.index();
1505     old_mems.pop();
1506     assert(old_cnt == old_mem->outcnt(), "old mem could be lost");
1507   }
1508 #endif
1509 }
1510 
1511 bool ConnectionGraph::has_candidates(Compile *C) {
1512   // EA brings benefits only when the code has allocations and/or locks which
1513   // are represented by ideal Macro nodes.
1514   int cnt = C->macro_count();
1515   for( int i=0; i < cnt; i++ ) {
1516     Node *n = C->macro_node(i);
1517     if ( n->is_Allocate() )
1518       return true;
1519     if( n->is_Lock() ) {
1520       Node* obj = n->as_Lock()->obj_node()->uncast();
1521       if( !(obj->is_Parm() || obj->is_Con()) )
1522         return true;
1523     }
1524   }
1525   return false;
1526 }
1527 
1528 void ConnectionGraph::do_analysis(Compile *C, PhaseIterGVN *igvn) {
1529   // Add ConP#NULL and ConN#NULL nodes before ConnectionGraph construction
1530   // to create space for them in ConnectionGraph::_nodes[].
1531   Node* oop_null = igvn->zerocon(T_OBJECT);
1532   Node* noop_null = igvn->zerocon(T_NARROWOOP);
1533 
1534   ConnectionGraph* congraph = new(C->comp_arena()) ConnectionGraph(C, igvn);
1535   // Perform escape analysis
1536   if (congraph->compute_escape()) {
1537     // There are non escaping objects.
1538     C->set_congraph(congraph);
1539   }
1540 
1541   // Cleanup.
1542   if (oop_null->outcnt() == 0)
1543     igvn->hash_delete(oop_null);
1544   if (noop_null->outcnt() == 0)
1545     igvn->hash_delete(noop_null);
1546 }
1547 
1548 bool ConnectionGraph::compute_escape() {
1549   Compile* C = _compile;
1550 
1551   // 1. Populate Connection Graph (CG) with Ideal nodes.
1552 
1553   Unique_Node_List worklist_init;
1554   worklist_init.map(C->unique(), NULL);  // preallocate space
1555 
1556   // Initialize worklist
1557   if (C->root() != NULL) {
1558     worklist_init.push(C->root());
1559   }
1560 
1561   GrowableArray<Node*> alloc_worklist;
1562   GrowableArray<Node*> addp_worklist;
1563   GrowableArray<Node*> ptr_cmp_worklist;
1564   PhaseGVN* igvn = _igvn;
1565   bool has_allocations = false;
1566 
1567   // Push all useful nodes onto CG list and set their type.
1568   for( uint next = 0; next < worklist_init.size(); ++next ) {
1569     Node* n = worklist_init.at(next);
1570     record_for_escape_analysis(n, igvn);
1571     // Only allocations and java static calls results are checked
1572     // for an escape status. See process_call_result() below.
1573     if (n->is_Allocate() || n->is_CallStaticJava() &&
1574         ptnode_adr(n->_idx)->node_type() == PointsToNode::JavaObject) {
1575       has_allocations = true;
1576       if (n->is_Allocate())
1577         alloc_worklist.append(n);
1578     } else if(n->is_AddP()) {
1579       // Collect address nodes. Use them during stage 3 below
1580       // to build initial connection graph field edges.
1581       addp_worklist.append(n);
1582     } else if (n->is_MergeMem()) {
1583       // Collect all MergeMem nodes to add memory slices for
1584       // scalar replaceable objects in split_unique_types().
1585       _mergemem_worklist.append(n->as_MergeMem());
1586     } else if (OptimizePtrCompare && n->is_Cmp() &&
1587                (n->Opcode() == Op_CmpP || n->Opcode() == Op_CmpN)) {
1588       // Compare pointers nodes
1589       ptr_cmp_worklist.append(n);
1590     }
1591     for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
1592       Node* m = n->fast_out(i);   // Get user
1593       worklist_init.push(m);
1594     }
1595   }
1596 
1597   if (!has_allocations) {
1598     _collecting = false;
1599     return false; // Nothing to do.
1600   }
1601 
1602   // 2. First pass to create simple CG edges (doesn't require to walk CG).
1603   uint delayed_size = _delayed_worklist.size();
1604   for( uint next = 0; next < delayed_size; ++next ) {
1605     Node* n = _delayed_worklist.at(next);
1606     build_connection_graph(n, igvn);
1607   }
1608 
1609   // 3. Pass to create initial fields edges (JavaObject -F-> AddP)
1610   //    to reduce number of iterations during stage 4 below.
1611   uint addp_length = addp_worklist.length();
1612   for( uint next = 0; next < addp_length; ++next ) {
1613     Node* n = addp_worklist.at(next);
1614     Node* base = get_addp_base(n);
1615     if (base->is_Proj() && base->in(0)->is_Call())
1616       base = base->in(0);
1617     PointsToNode::NodeType nt = ptnode_adr(base->_idx)->node_type();
1618     if (nt == PointsToNode::JavaObject) {
1619       build_connection_graph(n, igvn);
1620     }
1621   }
1622 
1623   GrowableArray<int> cg_worklist;
1624   cg_worklist.append(_phantom_object);
1625   GrowableArray<uint>  worklist;
1626 
1627   // 4. Build Connection Graph which need
1628   //    to walk the connection graph.
1629   _progress = false;
1630   for (uint ni = 0; ni < nodes_size(); ni++) {
1631     PointsToNode* ptn = ptnode_adr(ni);
1632     Node *n = ptn->_node;
1633     if (n != NULL) { // Call, AddP, LoadP, StoreP
1634       build_connection_graph(n, igvn);
1635       if (ptn->node_type() != PointsToNode::UnknownType)
1636         cg_worklist.append(n->_idx); // Collect CG nodes
1637       if (!_processed.test(n->_idx))
1638         worklist.append(n->_idx); // Collect C/A/L/S nodes
1639     }
1640   }
1641 
1642   // After IGVN user nodes may have smaller _idx than
1643   // their inputs so they will be processed first in
1644   // previous loop. Because of that not all Graph
1645   // edges will be created. Walk over interesting
1646   // nodes again until no new edges are created.
1647   //
1648   // Normally only 1-3 passes needed to build
1649   // Connection Graph depending on graph complexity.
1650   // Observed 8 passes in jvm2008 compiler.compiler.
1651   // Set limit to 20 to catch situation when something
1652   // did go wrong and recompile the method without EA.
1653 
1654 #define CG_BUILD_ITER_LIMIT 20
1655 
1656   uint length = worklist.length();
1657   int iterations = 0;
1658   while(_progress && (iterations++ < CG_BUILD_ITER_LIMIT)) {
1659     _progress = false;
1660     for( uint next = 0; next < length; ++next ) {
1661       int ni = worklist.at(next);
1662       PointsToNode* ptn = ptnode_adr(ni);
1663       Node* n = ptn->_node;
1664       assert(n != NULL, "should be known node");
1665       build_connection_graph(n, igvn);
1666     }
1667   }
1668   if (iterations >= CG_BUILD_ITER_LIMIT) {
1669     assert(iterations < CG_BUILD_ITER_LIMIT,
1670            err_msg("infinite EA connection graph build with %d nodes and worklist size %d",
1671            nodes_size(), length));
1672     // Possible infinite build_connection_graph loop,
1673     // retry compilation without escape analysis.
1674     C->record_failure(C2Compiler::retry_no_escape_analysis());
1675     _collecting = false;
1676     return false;
1677   }
1678 #undef CG_BUILD_ITER_LIMIT
1679 
1680   Arena* arena = Thread::current()->resource_area();
1681   VectorSet visited(arena);
1682 
1683   // 5. Find fields initializing values for not escaped allocations
1684   uint alloc_length = alloc_worklist.length();
1685   for (uint next = 0; next < alloc_length; ++next) {
1686     Node* n = alloc_worklist.at(next);
1687     if (ptnode_adr(n->_idx)->escape_state() == PointsToNode::NoEscape) {
1688       find_init_values(n, &visited, igvn);
1689     }
1690   }
1691 
1692   worklist.clear();
1693 
1694   // 6. Remove deferred edges from the graph.
1695   uint cg_length = cg_worklist.length();
1696   for (uint next = 0; next < cg_length; ++next) {
1697     int ni = cg_worklist.at(next);
1698     PointsToNode* ptn = ptnode_adr(ni);
1699     PointsToNode::NodeType nt = ptn->node_type();
1700     if (nt == PointsToNode::LocalVar || nt == PointsToNode::Field) {
1701       remove_deferred(ni, &worklist, &visited);
1702     }
1703   }
1704 
1705   // 7. Adjust escape state of nonescaping objects.
1706   for (uint next = 0; next < addp_length; ++next) {
1707     Node* n = addp_worklist.at(next);
1708     adjust_escape_state(n);
1709   }
1710 
1711   // 8. Propagate escape states.
1712   worklist.clear();
1713 
1714   // mark all nodes reachable from GlobalEscape nodes
1715   (void)propagate_escape_state(&cg_worklist, &worklist, PointsToNode::GlobalEscape);
1716 
1717   // mark all nodes reachable from ArgEscape nodes
1718   bool has_non_escaping_obj = propagate_escape_state(&cg_worklist, &worklist, PointsToNode::ArgEscape);
1719 
1720   // push all NoEscape nodes on the worklist
1721   for( uint next = 0; next < cg_length; ++next ) {
1722     int nk = cg_worklist.at(next);
1723     if (ptnode_adr(nk)->escape_state() == PointsToNode::NoEscape)
1724       worklist.push(nk);
1725   }
1726   alloc_worklist.clear();
1727   // mark all nodes reachable from NoEscape nodes
1728   while(worklist.length() > 0) {
1729     uint nk = worklist.pop();
1730     PointsToNode* ptn = ptnode_adr(nk);
1731     if (ptn->node_type() == PointsToNode::JavaObject &&
1732         !(nk == _noop_null || nk == _oop_null))
1733       has_non_escaping_obj = true; // Non Escape
1734     Node* n = ptn->_node;
1735     bool scalar_replaceable = ptn->scalar_replaceable();
1736     if (n->is_Allocate() && scalar_replaceable) {
1737       // Push scalar replaceable allocations on alloc_worklist
1738       // for processing in split_unique_types(). Note,
1739       // following code may change scalar_replaceable value.
1740       alloc_worklist.append(n);
1741     }
1742     uint e_cnt = ptn->edge_count();
1743     for (uint ei = 0; ei < e_cnt; ei++) {
1744       uint npi = ptn->edge_target(ei);
1745       PointsToNode *np = ptnode_adr(npi);
1746       if (np->escape_state() < PointsToNode::NoEscape) {
1747         set_escape_state(npi, PointsToNode::NoEscape);
1748         if (!scalar_replaceable) {
1749           np->set_scalar_replaceable(false);
1750         }
1751         worklist.push(npi);
1752       } else if (np->scalar_replaceable() && !scalar_replaceable) {
1753         // Propagate scalar_replaceable value.
1754         np->set_scalar_replaceable(false);
1755         worklist.push(npi);
1756       }
1757     }
1758   }
1759 
1760   _collecting = false;
1761   assert(C->unique() == nodes_size(), "there should be no new ideal nodes during ConnectionGraph build");
1762 
1763   assert(ptnode_adr(_oop_null)->escape_state() == PointsToNode::NoEscape, "sanity");
1764   if (UseCompressedOops) {
1765     assert(ptnode_adr(_noop_null)->escape_state() == PointsToNode::NoEscape, "sanity");
1766   }
1767 
1768   if (EliminateLocks && has_non_escaping_obj) {
1769     // Mark locks before changing ideal graph.
1770     int cnt = C->macro_count();
1771     for( int i=0; i < cnt; i++ ) {
1772       Node *n = C->macro_node(i);
1773       if (n->is_AbstractLock()) { // Lock and Unlock nodes
1774         AbstractLockNode* alock = n->as_AbstractLock();
1775         if (!alock->is_eliminated() || alock->is_coarsened()) {
1776           PointsToNode::EscapeState es = escape_state(alock->obj_node());
1777           assert(es != PointsToNode::UnknownEscape, "should know");
1778           if (es != PointsToNode::UnknownEscape && es != PointsToNode::GlobalEscape) {
1779             if (!alock->is_eliminated()) {
1780               // Mark it eliminated to update any counters
1781               alock->set_eliminated();
1782             } else {
1783               // The lock could be marked eliminated by lock coarsening
1784               // code during first IGVN before EA. Clear coarsened flag
1785               // to eliminate all associated locks/unlocks and relock
1786               // during deoptimization.
1787               alock->clear_coarsened();
1788             }
1789           }
1790         }
1791       }
1792     }
1793   }
1794 
1795   if (OptimizePtrCompare && has_non_escaping_obj) {
1796     // Add ConI(#CC_GT) and ConI(#CC_EQ).
1797     _pcmp_neq = igvn->makecon(TypeInt::CC_GT);
1798     _pcmp_eq = igvn->makecon(TypeInt::CC_EQ);
1799     // Optimize objects compare.
1800     while (ptr_cmp_worklist.length() != 0) {
1801       Node *n = ptr_cmp_worklist.pop();
1802       Node *res = optimize_ptr_compare(n);
1803       if (res != NULL) {
1804 #ifndef PRODUCT
1805         if (PrintOptimizePtrCompare) {
1806           tty->print_cr("++++ Replaced: %d %s(%d,%d) --> %s", n->_idx, (n->Opcode() == Op_CmpP ? "CmpP" : "CmpN"), n->in(1)->_idx, n->in(2)->_idx, (res == _pcmp_eq ? "EQ" : "NotEQ"));
1807           if (Verbose) {
1808             n->dump(1);
1809           }
1810         }
1811 #endif
1812         _igvn->replace_node(n, res);
1813       }
1814     }
1815     // cleanup
1816     if (_pcmp_neq->outcnt() == 0)
1817       igvn->hash_delete(_pcmp_neq);
1818     if (_pcmp_eq->outcnt()  == 0)
1819       igvn->hash_delete(_pcmp_eq);
1820   }
1821 
1822 #ifndef PRODUCT
1823   if (PrintEscapeAnalysis) {
1824     dump(); // Dump ConnectionGraph
1825   }
1826 #endif
1827 
1828   bool has_scalar_replaceable_candidates = false;
1829   alloc_length = alloc_worklist.length();
1830   for (uint next = 0; next < alloc_length; ++next) {
1831     Node* n = alloc_worklist.at(next);
1832     PointsToNode* ptn = ptnode_adr(n->_idx);
1833     assert(ptn->escape_state() == PointsToNode::NoEscape, "sanity");
1834     if (ptn->scalar_replaceable()) {
1835       has_scalar_replaceable_candidates = true;
1836       break;
1837     }
1838   }
1839 
1840   if ( has_scalar_replaceable_candidates &&
1841        C->AliasLevel() >= 3 && EliminateAllocations ) {
1842 
1843     // Now use the escape information to create unique types for
1844     // scalar replaceable objects.
1845     split_unique_types(alloc_worklist);
1846 
1847     if (C->failing())  return false;
1848 
1849     C->print_method("After Escape Analysis", 2);
1850 
1851 #ifdef ASSERT
1852   } else if (Verbose && (PrintEscapeAnalysis || PrintEliminateAllocations)) {
1853     tty->print("=== No allocations eliminated for ");
1854     C->method()->print_short_name();
1855     if(!EliminateAllocations) {
1856       tty->print(" since EliminateAllocations is off ===");
1857     } else if(!has_scalar_replaceable_candidates) {
1858       tty->print(" since there are no scalar replaceable candidates ===");
1859     } else if(C->AliasLevel() < 3) {
1860       tty->print(" since AliasLevel < 3 ===");
1861     }
1862     tty->cr();
1863 #endif
1864   }
1865   return has_non_escaping_obj;
1866 }
1867 
1868 // Find fields initializing values for allocations.
1869 void ConnectionGraph::find_init_values(Node* alloc, VectorSet* visited, PhaseTransform* phase) {
1870   assert(alloc->is_Allocate(), "Should be called for Allocate nodes only");
1871   PointsToNode* pta = ptnode_adr(alloc->_idx);
1872   assert(pta->escape_state() == PointsToNode::NoEscape, "Not escaped Allocate nodes only");
1873   InitializeNode* ini = alloc->as_Allocate()->initialization();
1874 
1875   Compile* C = _compile;
1876   visited->Reset();
1877   // Check if a oop field's initializing value is recorded and add
1878   // a corresponding NULL field's value if it is not recorded.
1879   // Connection Graph does not record a default initialization by NULL
1880   // captured by Initialize node.
1881   //
1882   uint ae_cnt = pta->edge_count();
1883   for (uint ei = 0; ei < ae_cnt; ei++) {
1884     uint nidx = pta->edge_target(ei); // Field (AddP)
1885     PointsToNode* ptn = ptnode_adr(nidx);
1886     assert(ptn->_node->is_AddP(), "Should be AddP nodes only");
1887     int offset = ptn->offset();
1888     if (offset != Type::OffsetBot &&
1889         offset != oopDesc::klass_offset_in_bytes() &&
1890         !visited->test_set(offset)) {
1891 
1892       // Check only oop fields.
1893       const Type* adr_type = ptn->_node->as_AddP()->bottom_type();
1894       BasicType basic_field_type = T_INT;
1895       if (adr_type->isa_instptr()) {
1896         ciField* field = C->alias_type(adr_type->isa_instptr())->field();
1897         if (field != NULL) {
1898           basic_field_type = field->layout_type();
1899         } else {
1900           // Ignore non field load (for example, klass load)
1901         }
1902       } else if (adr_type->isa_aryptr()) {
1903         if (offset != arrayOopDesc::length_offset_in_bytes()) {
1904           const Type* elemtype = adr_type->isa_aryptr()->elem();
1905           basic_field_type = elemtype->array_element_basic_type();
1906         } else {
1907           // Ignore array length load
1908         }
1909 #ifdef ASSERT
1910       } else {
1911         // Raw pointers are used for initializing stores so skip it
1912         // since it should be recorded already
1913         Node* base = get_addp_base(ptn->_node);
1914         assert(adr_type->isa_rawptr() && base->is_Proj() &&
1915                (base->in(0) == alloc),"unexpected pointer type");
1916 #endif
1917       }
1918       if (basic_field_type == T_OBJECT ||
1919           basic_field_type == T_NARROWOOP ||
1920           basic_field_type == T_ARRAY) {
1921         Node* value = NULL;
1922         if (ini != NULL) {
1923           BasicType ft = UseCompressedOops ? T_NARROWOOP : T_OBJECT;
1924           Node* store = ini->find_captured_store(offset, type2aelembytes(ft), phase);
1925           if (store != NULL && store->is_Store()) {
1926             value = store->in(MemNode::ValueIn);
1927           } else if (ptn->edge_count() > 0) { // Are there oop stores?
1928             // Check for a store which follows allocation without branches.
1929             // For example, a volatile field store is not collected
1930             // by Initialize node. TODO: it would be nice to use idom() here.
1931             //
1932             // Search all references to the same field which use different
1933             // AddP nodes, for example, in the next case:
1934             //
1935             //    Point p[] = new Point[1];
1936             //    if ( x ) { p[0] = new Point(); p[0].x = x; }
1937             //    if ( p[0] != null ) { y = p[0].x; } // has CastPP
1938             //
1939             for (uint next = ei; (next < ae_cnt) && (value == NULL); next++) {
1940               uint fpi = pta->edge_target(next); // Field (AddP)
1941               PointsToNode *ptf = ptnode_adr(fpi);
1942               if (ptf->offset() == offset) {
1943                 Node* nf = ptf->_node;
1944                 for (DUIterator_Fast imax, i = nf->fast_outs(imax); i < imax; i++) {
1945                   store = nf->fast_out(i);
1946                   if (store->is_Store() && store->in(0) != NULL) {
1947                     Node* ctrl = store->in(0);
1948                     while(!(ctrl == ini || ctrl == alloc || ctrl == NULL ||
1949                             ctrl == C->root() || ctrl == C->top() || ctrl->is_Region() ||
1950                             ctrl->is_IfTrue() || ctrl->is_IfFalse())) {
1951                        ctrl = ctrl->in(0);
1952                     }
1953                     if (ctrl == ini || ctrl == alloc) {
1954                       value = store->in(MemNode::ValueIn);
1955                       break;
1956                     }
1957                   }
1958                 }
1959               }
1960             }
1961           }
1962         }
1963         if (value == NULL || value != ptnode_adr(value->_idx)->_node) {
1964           // A field's initializing value was not recorded. Add NULL.
1965           uint null_idx = UseCompressedOops ? _noop_null : _oop_null;
1966           add_edge_from_fields(alloc->_idx, null_idx, offset);
1967         }
1968       }
1969     }
1970   }
1971 }
1972 
1973 // Adjust escape state after Connection Graph is built.
1974 void ConnectionGraph::adjust_escape_state(Node* n) {
1975   PointsToNode* ptn = ptnode_adr(n->_idx);
1976   assert(n->is_AddP(), "Should be called for AddP nodes only");
1977   // Search for objects which are not scalar replaceable
1978   // and mark them to propagate the state to referenced objects.
1979   //
1980 
1981   int offset = ptn->offset();
1982   Node* base = get_addp_base(n);
1983   VectorSet* ptset = PointsTo(base);
1984   int ptset_size = ptset->Size();
1985 
1986   // An object is not scalar replaceable if the field which may point
1987   // to it has unknown offset (unknown element of an array of objects).
1988   //
1989 
1990   if (offset == Type::OffsetBot) {
1991     uint e_cnt = ptn->edge_count();
1992     for (uint ei = 0; ei < e_cnt; ei++) {
1993       uint npi = ptn->edge_target(ei);
1994       ptnode_adr(npi)->set_scalar_replaceable(false);
1995     }
1996   }
1997 
1998   // Currently an object is not scalar replaceable if a LoadStore node
1999   // access its field since the field value is unknown after it.
2000   //
2001   bool has_LoadStore = false;
2002   for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
2003     Node *use = n->fast_out(i);
2004     if (use->is_LoadStore()) {
2005       has_LoadStore = true;
2006       break;
2007     }
2008   }
2009   // An object is not scalar replaceable if the address points
2010   // to unknown field (unknown element for arrays, offset is OffsetBot).
2011   //
2012   // Or the address may point to more then one object. This may produce
2013   // the false positive result (set not scalar replaceable)
2014   // since the flow-insensitive escape analysis can't separate
2015   // the case when stores overwrite the field's value from the case
2016   // when stores happened on different control branches.
2017   //
2018   // Note: it will disable scalar replacement in some cases:
2019   //
2020   //    Point p[] = new Point[1];
2021   //    p[0] = new Point(); // Will be not scalar replaced
2022   //
2023   // but it will save us from incorrect optimizations in next cases:
2024   //
2025   //    Point p[] = new Point[1];
2026   //    if ( x ) p[0] = new Point(); // Will be not scalar replaced
2027   //
2028   if (ptset_size > 1 || ptset_size != 0 &&
2029       (has_LoadStore || offset == Type::OffsetBot)) {
2030     for( VectorSetI j(ptset); j.test(); ++j ) {
2031       ptnode_adr(j.elem)->set_scalar_replaceable(false);
2032     }
2033   }
2034 }
2035 
2036 // Propagate escape states to referenced nodes.
2037 bool ConnectionGraph::propagate_escape_state(GrowableArray<int>* cg_worklist,
2038                                              GrowableArray<uint>* worklist,
2039                                              PointsToNode::EscapeState esc_state) {
2040   bool has_java_obj = false;
2041 
2042   // push all nodes with the same escape state on the worklist
2043   uint cg_length = cg_worklist->length();
2044   for (uint next = 0; next < cg_length; ++next) {
2045     int nk = cg_worklist->at(next);
2046     if (ptnode_adr(nk)->escape_state() == esc_state)
2047       worklist->push(nk);
2048   }
2049   // mark all reachable nodes
2050   while (worklist->length() > 0) {
2051     PointsToNode* ptn = ptnode_adr(worklist->pop());
2052     if (ptn->node_type() == PointsToNode::JavaObject) {
2053       has_java_obj = true;
2054     }
2055     uint e_cnt = ptn->edge_count();
2056     for (uint ei = 0; ei < e_cnt; ei++) {
2057       uint npi = ptn->edge_target(ei);
2058       PointsToNode *np = ptnode_adr(npi);
2059       if (np->escape_state() < esc_state) {
2060         set_escape_state(npi, esc_state);
2061         worklist->push(npi);
2062       }
2063     }
2064   }
2065   // Has not escaping java objects
2066   return has_java_obj && (esc_state < PointsToNode::GlobalEscape);
2067 }
2068 
2069 // Optimize objects compare.
2070 Node* ConnectionGraph::optimize_ptr_compare(Node* n) {
2071   assert(OptimizePtrCompare, "sanity");
2072   // Clone returned Set since PointsTo() returns pointer
2073   // to the same structure ConnectionGraph.pt_ptset.
2074   VectorSet ptset1 = *PointsTo(n->in(1));
2075   VectorSet ptset2 = *PointsTo(n->in(2));
2076 
2077   // Check simple cases first.
2078   if (ptset1.Size() == 1) {
2079     uint pt1 = ptset1.getelem();
2080     PointsToNode* ptn1 = ptnode_adr(pt1);
2081     if (ptn1->escape_state() == PointsToNode::NoEscape) {
2082       if (ptset2.Size() == 1 && ptset2.getelem() == pt1) {
2083         // Comparing the same not escaping object.
2084         return _pcmp_eq;
2085       }
2086       Node* obj = ptn1->_node;
2087       // Comparing not escaping allocation.
2088       if ((obj->is_Allocate() || obj->is_CallStaticJava()) &&
2089           !ptset2.test(pt1)) {
2090         return _pcmp_neq; // This includes nullness check.
2091       }
2092     }
2093   } else if (ptset2.Size() == 1) {
2094     uint pt2 = ptset2.getelem();
2095     PointsToNode* ptn2 = ptnode_adr(pt2);
2096     if (ptn2->escape_state() == PointsToNode::NoEscape) {
2097       Node* obj = ptn2->_node;
2098       // Comparing not escaping allocation.
2099       if ((obj->is_Allocate() || obj->is_CallStaticJava()) &&
2100           !ptset1.test(pt2)) {
2101         return _pcmp_neq; // This includes nullness check.
2102       }
2103     }
2104   }
2105 
2106   if (!ptset1.disjoint(ptset2)) {
2107     return NULL; // Sets are not disjoint
2108   }
2109 
2110   // Sets are disjoint.
2111   bool set1_has_unknown_ptr = ptset1.test(_phantom_object) != 0;
2112   bool set2_has_unknown_ptr = ptset2.test(_phantom_object) != 0;
2113   bool set1_has_null_ptr   = (ptset1.test(_oop_null) | ptset1.test(_noop_null)) != 0;
2114   bool set2_has_null_ptr   = (ptset2.test(_oop_null) | ptset2.test(_noop_null)) != 0;
2115 
2116   if (set1_has_unknown_ptr && set2_has_null_ptr ||
2117       set2_has_unknown_ptr && set1_has_null_ptr) {
2118     // Check nullness of unknown object.
2119     return NULL;
2120   }
2121 
2122   // Disjointness by itself is not sufficient since
2123   // alias analysis is not complete for escaped objects.
2124   // Disjoint sets are definitely unrelated only when
2125   // at least one set has only not escaping objects.
2126   if (!set1_has_unknown_ptr && !set1_has_null_ptr) {
2127     bool has_only_non_escaping_alloc = true;
2128     for (VectorSetI i(&ptset1); i.test(); ++i) {
2129       uint pt = i.elem;
2130       PointsToNode* ptn = ptnode_adr(pt);
2131       Node* obj = ptn->_node;
2132       if (ptn->escape_state() != PointsToNode::NoEscape ||
2133           !(obj->is_Allocate() || obj->is_CallStaticJava())) {
2134         has_only_non_escaping_alloc = false;
2135         break;
2136       }
2137     }
2138     if (has_only_non_escaping_alloc) {
2139       return _pcmp_neq;
2140     }
2141   }
2142   if (!set2_has_unknown_ptr && !set2_has_null_ptr) {
2143     bool has_only_non_escaping_alloc = true;
2144     for (VectorSetI i(&ptset2); i.test(); ++i) {
2145       uint pt = i.elem;
2146       PointsToNode* ptn = ptnode_adr(pt);
2147       Node* obj = ptn->_node;
2148       if (ptn->escape_state() != PointsToNode::NoEscape ||
2149           !(obj->is_Allocate() || obj->is_CallStaticJava())) {
2150         has_only_non_escaping_alloc = false;
2151         break;
2152       }
2153     }
2154     if (has_only_non_escaping_alloc) {
2155       return _pcmp_neq;
2156     }
2157   }
2158   return NULL;
2159 }
2160 
2161 void ConnectionGraph::process_call_arguments(CallNode *call, PhaseTransform *phase) {
2162 
2163     switch (call->Opcode()) {
2164 #ifdef ASSERT
2165     case Op_Allocate:
2166     case Op_AllocateArray:
2167     case Op_Lock:
2168     case Op_Unlock:
2169       assert(false, "should be done already");
2170       break;
2171 #endif
2172     case Op_CallLeaf:
2173     case Op_CallLeafNoFP:
2174     {
2175       // Stub calls, objects do not escape but they are not scale replaceable.
2176       // Adjust escape state for outgoing arguments.
2177       const TypeTuple * d = call->tf()->domain();
2178       for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
2179         const Type* at = d->field_at(i);
2180         Node *arg = call->in(i)->uncast();
2181         const Type *aat = phase->type(arg);
2182         if (!arg->is_top() && at->isa_ptr() && aat->isa_ptr() &&
2183             ptnode_adr(arg->_idx)->escape_state() < PointsToNode::ArgEscape) {
2184 
2185           assert(aat == Type::TOP || aat == TypePtr::NULL_PTR ||
2186                  aat->isa_ptr() != NULL, "expecting an Ptr");
2187 #ifdef ASSERT
2188           if (!(call->Opcode() == Op_CallLeafNoFP &&
2189                 call->as_CallLeaf()->_name != NULL &&
2190                 (strstr(call->as_CallLeaf()->_name, "arraycopy")  != 0) ||
2191                 call->as_CallLeaf()->_name != NULL &&
2192                 (strcmp(call->as_CallLeaf()->_name, "g1_wb_pre")  == 0 ||
2193                  strcmp(call->as_CallLeaf()->_name, "g1_wb_post") == 0 ))
2194           ) {
2195             call->dump();
2196             assert(false, "EA: unexpected CallLeaf");
2197           }
2198 #endif
2199           set_escape_state(arg->_idx, PointsToNode::ArgEscape);
2200           if (arg->is_AddP()) {
2201             //
2202             // The inline_native_clone() case when the arraycopy stub is called
2203             // after the allocation before Initialize and CheckCastPP nodes.
2204             //
2205             // Set AddP's base (Allocate) as not scalar replaceable since
2206             // pointer to the base (with offset) is passed as argument.
2207             //
2208             arg = get_addp_base(arg);
2209           }
2210           for( VectorSetI j(PointsTo(arg)); j.test(); ++j ) {
2211             uint pt = j.elem;
2212             set_escape_state(pt, PointsToNode::ArgEscape);
2213           }
2214         }
2215       }
2216       break;
2217     }
2218 
2219     case Op_CallStaticJava:
2220     // For a static call, we know exactly what method is being called.
2221     // Use bytecode estimator to record the call's escape affects
2222     {
2223       ciMethod *meth = call->as_CallJava()->method();
2224       BCEscapeAnalyzer *call_analyzer = (meth !=NULL) ? meth->get_bcea() : NULL;
2225       // fall-through if not a Java method or no analyzer information
2226       if (call_analyzer != NULL) {
2227         const TypeTuple * d = call->tf()->domain();
2228         bool copy_dependencies = false;
2229         for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
2230           const Type* at = d->field_at(i);
2231           int k = i - TypeFunc::Parms;
2232           Node *arg = call->in(i)->uncast();
2233 
2234           if (at->isa_oopptr() != NULL &&
2235               ptnode_adr(arg->_idx)->escape_state() < PointsToNode::GlobalEscape) {
2236 
2237             bool global_escapes = false;
2238             bool fields_escapes = false;
2239             if (!call_analyzer->is_arg_stack(k)) {
2240               // The argument global escapes, mark everything it could point to
2241               set_escape_state(arg->_idx, PointsToNode::GlobalEscape);
2242               global_escapes = true;
2243             } else {
2244               if (!call_analyzer->is_arg_local(k)) {
2245                 // The argument itself doesn't escape, but any fields might
2246                 fields_escapes = true;
2247               }
2248               set_escape_state(arg->_idx, PointsToNode::ArgEscape);
2249               copy_dependencies = true;
2250             }
2251 
2252             for( VectorSetI j(PointsTo(arg)); j.test(); ++j ) {
2253               uint pt = j.elem;
2254               if (global_escapes) {
2255                 //The argument global escapes, mark everything it could point to
2256                 set_escape_state(pt, PointsToNode::GlobalEscape);
2257               } else {
2258                 if (fields_escapes) {
2259                   // The argument itself doesn't escape, but any fields might
2260                   add_edge_from_fields(pt, _phantom_object, Type::OffsetBot);
2261                 }
2262                 set_escape_state(pt, PointsToNode::ArgEscape);
2263               }
2264             }
2265           }
2266         }
2267         if (copy_dependencies)
2268           call_analyzer->copy_dependencies(_compile->dependencies());
2269         break;
2270       }
2271     }
2272 
2273     default:
2274     // Fall-through here if not a Java method or no analyzer information
2275     // or some other type of call, assume the worst case: all arguments
2276     // globally escape.
2277     {
2278       // adjust escape state for  outgoing arguments
2279       const TypeTuple * d = call->tf()->domain();
2280       for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
2281         const Type* at = d->field_at(i);
2282         if (at->isa_oopptr() != NULL) {
2283           Node *arg = call->in(i)->uncast();
2284           set_escape_state(arg->_idx, PointsToNode::GlobalEscape);
2285           for( VectorSetI j(PointsTo(arg)); j.test(); ++j ) {
2286             uint pt = j.elem;
2287             set_escape_state(pt, PointsToNode::GlobalEscape);
2288           }
2289         }
2290       }
2291     }
2292   }
2293 }
2294 void ConnectionGraph::process_call_result(ProjNode *resproj, PhaseTransform *phase) {
2295   CallNode   *call = resproj->in(0)->as_Call();
2296   uint    call_idx = call->_idx;
2297   uint resproj_idx = resproj->_idx;
2298 
2299   switch (call->Opcode()) {
2300     case Op_Allocate:
2301     {
2302       Node *k = call->in(AllocateNode::KlassNode);
2303       const TypeKlassPtr *kt = k->bottom_type()->isa_klassptr();
2304       assert(kt != NULL, "TypeKlassPtr  required.");
2305       ciKlass* cik = kt->klass();
2306 
2307       PointsToNode::EscapeState es;
2308       uint edge_to;
2309       if (cik->is_subclass_of(_compile->env()->Thread_klass()) ||
2310          !cik->is_instance_klass() || // StressReflectiveCode
2311           cik->as_instance_klass()->has_finalizer()) {
2312         es = PointsToNode::GlobalEscape;
2313         edge_to = _phantom_object; // Could not be worse
2314       } else {
2315         es = PointsToNode::NoEscape;
2316         edge_to = call_idx;
2317         assert(ptnode_adr(call_idx)->scalar_replaceable(), "sanity");
2318       }
2319       set_escape_state(call_idx, es);
2320       add_pointsto_edge(resproj_idx, edge_to);
2321       _processed.set(resproj_idx);
2322       break;
2323     }
2324 
2325     case Op_AllocateArray:
2326     {
2327 
2328       Node *k = call->in(AllocateNode::KlassNode);
2329       const TypeKlassPtr *kt = k->bottom_type()->isa_klassptr();
2330       assert(kt != NULL, "TypeKlassPtr  required.");
2331       ciKlass* cik = kt->klass();
2332 
2333       PointsToNode::EscapeState es;
2334       uint edge_to;
2335       if (!cik->is_array_klass()) { // StressReflectiveCode
2336         es = PointsToNode::GlobalEscape;
2337         edge_to = _phantom_object;
2338       } else {
2339         es = PointsToNode::NoEscape;
2340         edge_to = call_idx;
2341         assert(ptnode_adr(call_idx)->scalar_replaceable(), "sanity");
2342         int length = call->in(AllocateNode::ALength)->find_int_con(-1);
2343         if (length < 0 || length > EliminateAllocationArraySizeLimit) {
2344           // Not scalar replaceable if the length is not constant or too big.
2345           ptnode_adr(call_idx)->set_scalar_replaceable(false);
2346         }
2347       }
2348       set_escape_state(call_idx, es);
2349       add_pointsto_edge(resproj_idx, edge_to);
2350       _processed.set(resproj_idx);
2351       break;
2352     }
2353 
2354     case Op_CallStaticJava:
2355     // For a static call, we know exactly what method is being called.
2356     // Use bytecode estimator to record whether the call's return value escapes
2357     {
2358       bool done = true;
2359       const TypeTuple *r = call->tf()->range();
2360       const Type* ret_type = NULL;
2361 
2362       if (r->cnt() > TypeFunc::Parms)
2363         ret_type = r->field_at(TypeFunc::Parms);
2364 
2365       // Note:  we use isa_ptr() instead of isa_oopptr()  here because the
2366       //        _multianewarray functions return a TypeRawPtr.
2367       if (ret_type == NULL || ret_type->isa_ptr() == NULL) {
2368         _processed.set(resproj_idx);
2369         break;  // doesn't return a pointer type
2370       }
2371       ciMethod *meth = call->as_CallJava()->method();
2372       const TypeTuple * d = call->tf()->domain();
2373       if (meth == NULL) {
2374         // not a Java method, assume global escape
2375         set_escape_state(call_idx, PointsToNode::GlobalEscape);
2376         add_pointsto_edge(resproj_idx, _phantom_object);
2377       } else {
2378         BCEscapeAnalyzer *call_analyzer = meth->get_bcea();
2379         bool copy_dependencies = false;
2380 
2381         if (call_analyzer->is_return_allocated()) {
2382           // Returns a newly allocated unescaped object, simply
2383           // update dependency information.
2384           // Mark it as NoEscape so that objects referenced by
2385           // it's fields will be marked as NoEscape at least.
2386           set_escape_state(call_idx, PointsToNode::NoEscape);
2387           ptnode_adr(call_idx)->set_scalar_replaceable(false);
2388           add_pointsto_edge(resproj_idx, call_idx);
2389           copy_dependencies = true;
2390         } else if (call_analyzer->is_return_local()) {
2391           // determine whether any arguments are returned
2392           set_escape_state(call_idx, PointsToNode::ArgEscape);
2393           bool ret_arg = false;
2394           for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
2395             const Type* at = d->field_at(i);
2396 
2397             if (at->isa_oopptr() != NULL) {
2398               Node *arg = call->in(i)->uncast();
2399 
2400               if (call_analyzer->is_arg_returned(i - TypeFunc::Parms)) {
2401                 ret_arg = true;
2402                 PointsToNode *arg_esp = ptnode_adr(arg->_idx);
2403                 if (arg_esp->node_type() == PointsToNode::UnknownType)
2404                   done = false;
2405                 else if (arg_esp->node_type() == PointsToNode::JavaObject)
2406                   add_pointsto_edge(resproj_idx, arg->_idx);
2407                 else
2408                   add_deferred_edge(resproj_idx, arg->_idx);
2409               }
2410             }
2411           }
2412           if (done && !ret_arg) {
2413             // Returns unknown object.
2414             set_escape_state(call_idx, PointsToNode::GlobalEscape);
2415             add_pointsto_edge(resproj_idx, _phantom_object);
2416           }
2417           if (done) {
2418             copy_dependencies = true;
2419           }
2420         } else {
2421           set_escape_state(call_idx, PointsToNode::GlobalEscape);
2422           add_pointsto_edge(resproj_idx, _phantom_object);
2423         }
2424         if (copy_dependencies)
2425           call_analyzer->copy_dependencies(_compile->dependencies());
2426       }
2427       if (done)
2428         _processed.set(resproj_idx);
2429       break;
2430     }
2431 
2432     default:
2433     // Some other type of call, assume the worst case that the
2434     // returned value, if any, globally escapes.
2435     {
2436       const TypeTuple *r = call->tf()->range();
2437       if (r->cnt() > TypeFunc::Parms) {
2438         const Type* ret_type = r->field_at(TypeFunc::Parms);
2439 
2440         // Note:  we use isa_ptr() instead of isa_oopptr()  here because the
2441         //        _multianewarray functions return a TypeRawPtr.
2442         if (ret_type->isa_ptr() != NULL) {
2443           set_escape_state(call_idx, PointsToNode::GlobalEscape);
2444           add_pointsto_edge(resproj_idx, _phantom_object);
2445         }
2446       }
2447       _processed.set(resproj_idx);
2448     }
2449   }
2450 }
2451 
2452 // Populate Connection Graph with Ideal nodes and create simple
2453 // connection graph edges (do not need to check the node_type of inputs
2454 // or to call PointsTo() to walk the connection graph).
2455 void ConnectionGraph::record_for_escape_analysis(Node *n, PhaseTransform *phase) {
2456   if (_processed.test(n->_idx))
2457     return; // No need to redefine node's state.
2458 
2459   if (n->is_Call()) {
2460     // Arguments to allocation and locking don't escape.
2461     if (n->is_Allocate()) {
2462       add_node(n, PointsToNode::JavaObject, PointsToNode::UnknownEscape, true);
2463       record_for_optimizer(n);
2464     } else if (n->is_Lock() || n->is_Unlock()) {
2465       // Put Lock and Unlock nodes on IGVN worklist to process them during
2466       // the first IGVN optimization when escape information is still available.
2467       record_for_optimizer(n);
2468       _processed.set(n->_idx);
2469     } else {
2470       // Don't mark as processed since call's arguments have to be processed.
2471       PointsToNode::NodeType nt = PointsToNode::UnknownType;
2472       PointsToNode::EscapeState es = PointsToNode::UnknownEscape;
2473 
2474       // Check if a call returns an object.
2475       const TypeTuple *r = n->as_Call()->tf()->range();
2476       if (r->cnt() > TypeFunc::Parms &&
2477           r->field_at(TypeFunc::Parms)->isa_ptr() &&
2478           n->as_Call()->proj_out(TypeFunc::Parms) != NULL) {
2479         nt = PointsToNode::JavaObject;
2480         if (!n->is_CallStaticJava()) {
2481           // Since the called mathod is statically unknown assume
2482           // the worst case that the returned value globally escapes.
2483           es = PointsToNode::GlobalEscape;
2484         }
2485       }
2486       add_node(n, nt, es, false);
2487     }
2488     return;
2489   }
2490 
2491   // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
2492   // ThreadLocal has RawPrt type.
2493   switch (n->Opcode()) {
2494     case Op_AddP:
2495     {
2496       add_node(n, PointsToNode::Field, PointsToNode::UnknownEscape, false);
2497       break;
2498     }
2499     case Op_CastX2P:
2500     { // "Unsafe" memory access.
2501       add_node(n, PointsToNode::JavaObject, PointsToNode::GlobalEscape, true);
2502       break;
2503     }
2504     case Op_CastPP:
2505     case Op_CheckCastPP:
2506     case Op_EncodeP:
2507     case Op_DecodeN:
2508     {
2509       add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false);
2510       int ti = n->in(1)->_idx;
2511       PointsToNode::NodeType nt = ptnode_adr(ti)->node_type();
2512       if (nt == PointsToNode::UnknownType) {
2513         _delayed_worklist.push(n); // Process it later.
2514         break;
2515       } else if (nt == PointsToNode::JavaObject) {
2516         add_pointsto_edge(n->_idx, ti);
2517       } else {
2518         add_deferred_edge(n->_idx, ti);
2519       }
2520       _processed.set(n->_idx);
2521       break;
2522     }
2523     case Op_ConP:
2524     {
2525       // assume all pointer constants globally escape except for null
2526       PointsToNode::EscapeState es;
2527       if (phase->type(n) == TypePtr::NULL_PTR)
2528         es = PointsToNode::NoEscape;
2529       else
2530         es = PointsToNode::GlobalEscape;
2531 
2532       add_node(n, PointsToNode::JavaObject, es, true);
2533       break;
2534     }
2535     case Op_ConN:
2536     {
2537       // assume all narrow oop constants globally escape except for null
2538       PointsToNode::EscapeState es;
2539       if (phase->type(n) == TypeNarrowOop::NULL_PTR)
2540         es = PointsToNode::NoEscape;
2541       else
2542         es = PointsToNode::GlobalEscape;
2543 
2544       add_node(n, PointsToNode::JavaObject, es, true);
2545       break;
2546     }
2547     case Op_CreateEx:
2548     {
2549       // assume that all exception objects globally escape
2550       add_node(n, PointsToNode::JavaObject, PointsToNode::GlobalEscape, true);
2551       break;
2552     }
2553     case Op_LoadKlass:
2554     case Op_LoadNKlass:
2555     {
2556       add_node(n, PointsToNode::JavaObject, PointsToNode::GlobalEscape, true);
2557       break;
2558     }
2559     case Op_LoadP:
2560     case Op_LoadN:
2561     {
2562       const Type *t = phase->type(n);
2563       if (t->make_ptr() == NULL) {
2564         _processed.set(n->_idx);
2565         return;
2566       }
2567       add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false);
2568       break;
2569     }
2570     case Op_Parm:
2571     {
2572       _processed.set(n->_idx); // No need to redefine it state.
2573       uint con = n->as_Proj()->_con;
2574       if (con < TypeFunc::Parms)
2575         return;
2576       const Type *t = n->in(0)->as_Start()->_domain->field_at(con);
2577       if (t->isa_ptr() == NULL)
2578         return;
2579       // We have to assume all input parameters globally escape
2580       // (Note: passing 'false' since _processed is already set).
2581       add_node(n, PointsToNode::JavaObject, PointsToNode::GlobalEscape, false);
2582       break;
2583     }
2584     case Op_PartialSubtypeCheck:
2585     { // Produces Null or notNull and is used in CmpP.
2586       add_node(n, PointsToNode::JavaObject, PointsToNode::ArgEscape, true);
2587       break;
2588     }
2589     case Op_Phi:
2590     {
2591       const Type *t = n->as_Phi()->type();
2592       if (t->make_ptr() == NULL) {
2593         // nothing to do if not an oop or narrow oop
2594         _processed.set(n->_idx);
2595         return;
2596       }
2597       add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false);
2598       uint i;
2599       for (i = 1; i < n->req() ; i++) {
2600         Node* in = n->in(i);
2601         if (in == NULL)
2602           continue;  // ignore NULL
2603         in = in->uncast();
2604         if (in->is_top() || in == n)
2605           continue;  // ignore top or inputs which go back this node
2606         int ti = in->_idx;
2607         PointsToNode::NodeType nt = ptnode_adr(ti)->node_type();
2608         if (nt == PointsToNode::UnknownType) {
2609           break;
2610         } else if (nt == PointsToNode::JavaObject) {
2611           add_pointsto_edge(n->_idx, ti);
2612         } else {
2613           add_deferred_edge(n->_idx, ti);
2614         }
2615       }
2616       if (i >= n->req())
2617         _processed.set(n->_idx);
2618       else
2619         _delayed_worklist.push(n);
2620       break;
2621     }
2622     case Op_Proj:
2623     {
2624       // we are only interested in the oop result projection from a call
2625       if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() ) {
2626         const TypeTuple *r = n->in(0)->as_Call()->tf()->range();
2627         assert(r->cnt() > TypeFunc::Parms, "sanity");
2628         if (r->field_at(TypeFunc::Parms)->isa_ptr() != NULL) {
2629           add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false);
2630           int ti = n->in(0)->_idx;
2631           // The call may not be registered yet (since not all its inputs are registered)
2632           // if this is the projection from backbranch edge of Phi.
2633           if (ptnode_adr(ti)->node_type() != PointsToNode::UnknownType) {
2634             process_call_result(n->as_Proj(), phase);
2635           }
2636           if (!_processed.test(n->_idx)) {
2637             // The call's result may need to be processed later if the call
2638             // returns it's argument and the argument is not processed yet.
2639             _delayed_worklist.push(n);
2640           }
2641           break;
2642         }
2643       }
2644       _processed.set(n->_idx);
2645       break;
2646     }
2647     case Op_Return:
2648     {
2649       if( n->req() > TypeFunc::Parms &&
2650           phase->type(n->in(TypeFunc::Parms))->isa_oopptr() ) {
2651         // Treat Return value as LocalVar with GlobalEscape escape state.
2652         add_node(n, PointsToNode::LocalVar, PointsToNode::GlobalEscape, false);
2653         int ti = n->in(TypeFunc::Parms)->_idx;
2654         PointsToNode::NodeType nt = ptnode_adr(ti)->node_type();
2655         if (nt == PointsToNode::UnknownType) {
2656           _delayed_worklist.push(n); // Process it later.
2657           break;
2658         } else if (nt == PointsToNode::JavaObject) {
2659           add_pointsto_edge(n->_idx, ti);
2660         } else {
2661           add_deferred_edge(n->_idx, ti);
2662         }
2663       }
2664       _processed.set(n->_idx);
2665       break;
2666     }
2667     case Op_StoreP:
2668     case Op_StoreN:
2669     {
2670       const Type *adr_type = phase->type(n->in(MemNode::Address));
2671       adr_type = adr_type->make_ptr();
2672       if (adr_type->isa_oopptr()) {
2673         add_node(n, PointsToNode::UnknownType, PointsToNode::UnknownEscape, false);
2674       } else {
2675         Node* adr = n->in(MemNode::Address);
2676         if (adr->is_AddP() && phase->type(adr) == TypeRawPtr::NOTNULL &&
2677             adr->in(AddPNode::Address)->is_Proj() &&
2678             adr->in(AddPNode::Address)->in(0)->is_Allocate()) {
2679           add_node(n, PointsToNode::UnknownType, PointsToNode::UnknownEscape, false);
2680           // We are computing a raw address for a store captured
2681           // by an Initialize compute an appropriate address type.
2682           int offs = (int)phase->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
2683           assert(offs != Type::OffsetBot, "offset must be a constant");
2684         } else {
2685           _processed.set(n->_idx);
2686           return;
2687         }
2688       }
2689       break;
2690     }
2691     case Op_StorePConditional:
2692     case Op_CompareAndSwapP:
2693     case Op_CompareAndSwapN:
2694     {
2695       const Type *adr_type = phase->type(n->in(MemNode::Address));
2696       adr_type = adr_type->make_ptr();
2697       if (adr_type->isa_oopptr()) {
2698         add_node(n, PointsToNode::UnknownType, PointsToNode::UnknownEscape, false);
2699       } else {
2700         _processed.set(n->_idx);
2701         return;
2702       }
2703       break;
2704     }
2705     case Op_AryEq:
2706     case Op_StrComp:
2707     case Op_StrEquals:
2708     case Op_StrIndexOf:
2709     {
2710       // char[] arrays passed to string intrinsics are not scalar replaceable.
2711       add_node(n, PointsToNode::UnknownType, PointsToNode::UnknownEscape, false);
2712       break;
2713     }
2714     case Op_ThreadLocal:
2715     {
2716       add_node(n, PointsToNode::JavaObject, PointsToNode::ArgEscape, true);
2717       break;
2718     }
2719     default:
2720       ;
2721       // nothing to do
2722   }
2723   return;
2724 }
2725 
2726 void ConnectionGraph::build_connection_graph(Node *n, PhaseTransform *phase) {
2727   uint n_idx = n->_idx;
2728   assert(ptnode_adr(n_idx)->_node != NULL, "node should be registered");
2729 
2730   // Don't set processed bit for AddP, LoadP, StoreP since
2731   // they may need more then one pass to process.
2732   // Also don't mark as processed Call nodes since their
2733   // arguments may need more then one pass to process.
2734   if (_processed.test(n_idx))
2735     return; // No need to redefine node's state.
2736 
2737   if (n->is_Call()) {
2738     CallNode *call = n->as_Call();
2739     process_call_arguments(call, phase);
2740     return;
2741   }
2742 
2743   switch (n->Opcode()) {
2744     case Op_AddP:
2745     {
2746       Node *base = get_addp_base(n);
2747       int offset = address_offset(n, phase);
2748       // Create a field edge to this node from everything base could point to.
2749       for( VectorSetI i(PointsTo(base)); i.test(); ++i ) {
2750         uint pt = i.elem;
2751         add_field_edge(pt, n_idx, offset);
2752       }
2753       break;
2754     }
2755     case Op_CastX2P:
2756     {
2757       assert(false, "Op_CastX2P");
2758       break;
2759     }
2760     case Op_CastPP:
2761     case Op_CheckCastPP:
2762     case Op_EncodeP:
2763     case Op_DecodeN:
2764     {
2765       int ti = n->in(1)->_idx;
2766       assert(ptnode_adr(ti)->node_type() != PointsToNode::UnknownType, "all nodes should be registered");
2767       if (ptnode_adr(ti)->node_type() == PointsToNode::JavaObject) {
2768         add_pointsto_edge(n_idx, ti);
2769       } else {
2770         add_deferred_edge(n_idx, ti);
2771       }
2772       _processed.set(n_idx);
2773       break;
2774     }
2775     case Op_ConP:
2776     {
2777       assert(false, "Op_ConP");
2778       break;
2779     }
2780     case Op_ConN:
2781     {
2782       assert(false, "Op_ConN");
2783       break;
2784     }
2785     case Op_CreateEx:
2786     {
2787       assert(false, "Op_CreateEx");
2788       break;
2789     }
2790     case Op_LoadKlass:
2791     case Op_LoadNKlass:
2792     {
2793       assert(false, "Op_LoadKlass");
2794       break;
2795     }
2796     case Op_LoadP:
2797     case Op_LoadN:
2798     {
2799       const Type *t = phase->type(n);
2800 #ifdef ASSERT
2801       if (t->make_ptr() == NULL)
2802         assert(false, "Op_LoadP");
2803 #endif
2804 
2805       Node* adr = n->in(MemNode::Address)->uncast();
2806       Node* adr_base;
2807       if (adr->is_AddP()) {
2808         adr_base = get_addp_base(adr);
2809       } else {
2810         adr_base = adr;
2811       }
2812 
2813       // For everything "adr_base" could point to, create a deferred edge from
2814       // this node to each field with the same offset.
2815       int offset = address_offset(adr, phase);
2816       for( VectorSetI i(PointsTo(adr_base)); i.test(); ++i ) {
2817         uint pt = i.elem;
2818         if (adr->is_AddP()) {
2819           // Add field edge if it is missing.
2820           add_field_edge(pt, adr->_idx, offset);
2821         }
2822         add_deferred_edge_to_fields(n_idx, pt, offset);
2823       }
2824       break;
2825     }
2826     case Op_Parm:
2827     {
2828       assert(false, "Op_Parm");
2829       break;
2830     }
2831     case Op_PartialSubtypeCheck:
2832     {
2833       assert(false, "Op_PartialSubtypeCheck");
2834       break;
2835     }
2836     case Op_Phi:
2837     {
2838 #ifdef ASSERT
2839       const Type *t = n->as_Phi()->type();
2840       if (t->make_ptr() == NULL)
2841         assert(false, "Op_Phi");
2842 #endif
2843       for (uint i = 1; i < n->req() ; i++) {
2844         Node* in = n->in(i);
2845         if (in == NULL)
2846           continue;  // ignore NULL
2847         in = in->uncast();
2848         if (in->is_top() || in == n)
2849           continue;  // ignore top or inputs which go back this node
2850         int ti = in->_idx;
2851         PointsToNode::NodeType nt = ptnode_adr(ti)->node_type();
2852         assert(nt != PointsToNode::UnknownType, "all nodes should be known");
2853         if (nt == PointsToNode::JavaObject) {
2854           add_pointsto_edge(n_idx, ti);
2855         } else {
2856           add_deferred_edge(n_idx, ti);
2857         }
2858       }
2859       _processed.set(n_idx);
2860       break;
2861     }
2862     case Op_Proj:
2863     {
2864       // we are only interested in the oop result projection from a call
2865       if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() ) {
2866         assert(ptnode_adr(n->in(0)->_idx)->node_type() != PointsToNode::UnknownType,
2867                "all nodes should be registered");
2868         const TypeTuple *r = n->in(0)->as_Call()->tf()->range();
2869         assert(r->cnt() > TypeFunc::Parms, "sanity");
2870         if (r->field_at(TypeFunc::Parms)->isa_ptr() != NULL) {
2871           process_call_result(n->as_Proj(), phase);
2872           assert(_processed.test(n_idx), "all call results should be processed");
2873           break;
2874         }
2875       }
2876       assert(false, "Op_Proj");
2877       break;
2878     }
2879     case Op_Return:
2880     {
2881 #ifdef ASSERT
2882       if( n->req() <= TypeFunc::Parms ||
2883           !phase->type(n->in(TypeFunc::Parms))->isa_oopptr() ) {
2884         assert(false, "Op_Return");
2885       }
2886 #endif
2887       int ti = n->in(TypeFunc::Parms)->_idx;
2888       assert(ptnode_adr(ti)->node_type() != PointsToNode::UnknownType, "node should be registered");
2889       if (ptnode_adr(ti)->node_type() == PointsToNode::JavaObject) {
2890         add_pointsto_edge(n_idx, ti);
2891       } else {
2892         add_deferred_edge(n_idx, ti);
2893       }
2894       _processed.set(n_idx);
2895       break;
2896     }
2897     case Op_StoreP:
2898     case Op_StoreN:
2899     case Op_StorePConditional:
2900     case Op_CompareAndSwapP:
2901     case Op_CompareAndSwapN:
2902     {
2903       Node *adr = n->in(MemNode::Address);
2904       const Type *adr_type = phase->type(adr)->make_ptr();
2905 #ifdef ASSERT
2906       if (!adr_type->isa_oopptr())
2907         assert(phase->type(adr) == TypeRawPtr::NOTNULL, "Op_StoreP");
2908 #endif
2909 
2910       assert(adr->is_AddP(), "expecting an AddP");
2911       Node *adr_base = get_addp_base(adr);
2912       Node *val = n->in(MemNode::ValueIn)->uncast();
2913       int offset = address_offset(adr, phase);
2914       // For everything "adr_base" could point to, create a deferred edge
2915       // to "val" from each field with the same offset.
2916       for( VectorSetI i(PointsTo(adr_base)); i.test(); ++i ) {
2917         uint pt = i.elem;
2918         // Add field edge if it is missing.
2919         add_field_edge(pt, adr->_idx, offset);
2920         add_edge_from_fields(pt, val->_idx, offset);
2921       }
2922       break;
2923     }
2924     case Op_AryEq:
2925     case Op_StrComp:
2926     case Op_StrEquals:
2927     case Op_StrIndexOf:
2928     {
2929       // char[] arrays passed to string intrinsic do not escape but
2930       // they are not scalar replaceable. Adjust escape state for them.
2931       // Start from in(2) edge since in(1) is memory edge.
2932       for (uint i = 2; i < n->req(); i++) {
2933         Node* adr = n->in(i)->uncast();
2934         const Type *at = phase->type(adr);
2935         if (!adr->is_top() && at->isa_ptr()) {
2936           assert(at == Type::TOP || at == TypePtr::NULL_PTR ||
2937                  at->isa_ptr() != NULL, "expecting an Ptr");
2938           if (adr->is_AddP()) {
2939             adr = get_addp_base(adr);
2940           }
2941           // Mark as ArgEscape everything "adr" could point to.
2942           set_escape_state(adr->_idx, PointsToNode::ArgEscape);
2943         }
2944       }
2945       _processed.set(n_idx);
2946       break;
2947     }
2948     case Op_ThreadLocal:
2949     {
2950       assert(false, "Op_ThreadLocal");
2951       break;
2952     }
2953     default:
2954       // This method should be called only for EA specific nodes.
2955       ShouldNotReachHere();
2956   }
2957 }
2958 
2959 #ifndef PRODUCT
2960 void ConnectionGraph::dump() {
2961   bool first = true;
2962 
2963   uint size = nodes_size();
2964   for (uint ni = 0; ni < size; ni++) {
2965     PointsToNode *ptn = ptnode_adr(ni);
2966     PointsToNode::NodeType ptn_type = ptn->node_type();
2967 
2968     if (ptn_type != PointsToNode::JavaObject || ptn->_node == NULL)
2969       continue;
2970     PointsToNode::EscapeState es = escape_state(ptn->_node);
2971     if (ptn->_node->is_Allocate() && (es == PointsToNode::NoEscape || Verbose)) {
2972       if (first) {
2973         tty->cr();
2974         tty->print("======== Connection graph for ");
2975         _compile->method()->print_short_name();
2976         tty->cr();
2977         first = false;
2978       }
2979       tty->print("%6d ", ni);
2980       ptn->dump();
2981       // Print all locals which reference this allocation
2982       for (uint li = ni; li < size; li++) {
2983         PointsToNode *ptn_loc = ptnode_adr(li);
2984         PointsToNode::NodeType ptn_loc_type = ptn_loc->node_type();
2985         if ( ptn_loc_type == PointsToNode::LocalVar && ptn_loc->_node != NULL &&
2986              ptn_loc->edge_count() == 1 && ptn_loc->edge_target(0) == ni ) {
2987           ptnode_adr(li)->dump(false);
2988         }
2989       }
2990       if (Verbose) {
2991         // Print all fields which reference this allocation
2992         for (uint i = 0; i < ptn->edge_count(); i++) {
2993           uint ei = ptn->edge_target(i);
2994           ptnode_adr(ei)->dump(false);
2995         }
2996       }
2997       tty->cr();
2998     }
2999   }
3000 }
3001 #endif