vm/opto/memnode.cpp

/*
 * Copyright 1997-2007 Sun Microsystems, Inc.  All Rights Reserved.
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
 * CA 95054 USA or visit www.sun.com if you need additional information or
 * have any questions.
 *
 */

// Portions of code courtesy of Clifford Click

// Optimization - Graph Style

#include "incls/_precompiled.incl"
#include "incls/_memnode.cpp.incl"

static Node *step_through_mergemem(PhaseGVN *phase, MergeMemNode *mmem,  const TypePtr *tp, const TypePtr *adr_check, outputStream *st);

//=============================================================================
uint MemNode::size_of() const { return sizeof(*this); }

const TypePtr *MemNode::adr_type() const {
  Node* adr = in(Address);
  const TypePtr* cross_check = NULL;
  DEBUG_ONLY(cross_check = _adr_type);
  return calculate_adr_type(adr->bottom_type(), cross_check);
}

#ifndef PRODUCT
void MemNode::dump_spec(outputStream *st) const {
  if (in(Address) == NULL)  return; // node is dead
#ifndef ASSERT
  // fake the missing field
  const TypePtr* _adr_type = NULL;
  if (in(Address) != NULL)
    _adr_type = in(Address)->bottom_type()->isa_ptr();
#endif
  dump_adr_type(this, _adr_type, st);

  Compile* C = Compile::current();
  if( C->alias_type(_adr_type)->is_volatile() )
    st->print(" Volatile!");
}

void MemNode::dump_adr_type(const Node* mem, const TypePtr* adr_type, outputStream *st) {
  st->print(" @");
  if (adr_type == NULL) {
    st->print("NULL");
  } else {
    adr_type->dump_on(st);
    Compile* C = Compile::current();
    Compile::AliasType* atp = NULL;
    if (C->have_alias_type(adr_type))  atp = C->alias_type(adr_type);
    if (atp == NULL)
      st->print(", idx=?\?;");
    else if (atp->index() == Compile::AliasIdxBot)
      st->print(", idx=Bot;");
    else if (atp->index() == Compile::AliasIdxTop)
      st->print(", idx=Top;");
    else if (atp->index() == Compile::AliasIdxRaw)
      st->print(", idx=Raw;");
    else {
      ciField* field = atp->field();
      if (field) {
        st->print(", name=");
        field->print_name_on(st);
      }
      st->print(", idx=%d;", atp->index());
    }
  }
}

extern void print_alias_types();

#endif

Node *MemNode::optimize_simple_memory_chain(Node *mchain, const TypePtr *t_adr, PhaseGVN *phase) {
  const TypeOopPtr *tinst = t_adr->isa_oopptr();
  if (tinst == NULL || !tinst->is_instance_field())
    return mchain;  // don't try to optimize non-instance types
  uint instance_id = tinst->instance_id();
  Node *prev = NULL;
  Node *result = mchain;
  while (prev != result) {
    prev = result;
    // skip over a call which does not affect this memory slice
    if (result->is_Proj() && result->as_Proj()->_con == TypeFunc::Memory) {
      Node *proj_in = result->in(0);
      if (proj_in->is_Call()) {
        CallNode *call = proj_in->as_Call();
        if (!call->may_modify(t_adr, phase)) {
          result = call->in(TypeFunc::Memory);
        }
      } else if (proj_in->is_Initialize()) {
        AllocateNode* alloc = proj_in->as_Initialize()->allocation();
        // Stop if this is the initialization for the object instance which
        // which contains this memory slice, otherwise skip over it.
        if (alloc != NULL && alloc->_idx != instance_id) {
          result = proj_in->in(TypeFunc::Memory);
        }
      } else if (proj_in->is_MemBar()) {
        result = proj_in->in(TypeFunc::Memory);
      }
    } else if (result->is_MergeMem()) {
      result = step_through_mergemem(phase, result->as_MergeMem(), t_adr, NULL, tty);
    }
  }
  return result;
}

Node *MemNode::optimize_memory_chain(Node *mchain, const TypePtr *t_adr, PhaseGVN *phase) {
  const TypeOopPtr *t_oop = t_adr->isa_oopptr();
  bool is_instance = (t_oop != NULL) && t_oop->is_instance_field();
  PhaseIterGVN *igvn = phase->is_IterGVN();
  Node *result = mchain;
  result = optimize_simple_memory_chain(result, t_adr, phase);
  if (is_instance && igvn != NULL  && result->is_Phi()) {
    PhiNode *mphi = result->as_Phi();
    assert(mphi->bottom_type() == Type::MEMORY, "memory phi required");
    const TypePtr *t = mphi->adr_type();
    if (t == TypePtr::BOTTOM || t == TypeRawPtr::BOTTOM) {
      // clone the Phi with our address type
      result = mphi->split_out_instance(t_adr, igvn);
    } else {
      assert(phase->C->get_alias_index(t) == phase->C->get_alias_index(t_adr), "correct memory chain");
    }
  }
  return result;
}

static Node *step_through_mergemem(PhaseGVN *phase, MergeMemNode *mmem,  const TypePtr *tp, const TypePtr *adr_check, outputStream *st) {
  uint alias_idx = phase->C->get_alias_index(tp);
  Node *mem = mmem;
#ifdef ASSERT
  {
    // Check that current type is consistent with the alias index used during graph construction
    assert(alias_idx >= Compile::AliasIdxRaw, "must not be a bad alias_idx");
    bool consistent =  adr_check == NULL || adr_check->empty() ||
                       phase->C->must_alias(adr_check, alias_idx );
    // Sometimes dead array references collapse to a[-1], a[-2], or a[-3]
    if( !consistent && adr_check != NULL && !adr_check->empty() &&
           tp->isa_aryptr() &&    tp->offset() == Type::OffsetBot &&
        adr_check->isa_aryptr() && adr_check->offset() != Type::OffsetBot &&
        ( adr_check->offset() == arrayOopDesc::length_offset_in_bytes() ||
          adr_check->offset() == oopDesc::klass_offset_in_bytes() ||
          adr_check->offset() == oopDesc::mark_offset_in_bytes() ) ) {
      // don't assert if it is dead code.
      consistent = true;
    }
    if( !consistent ) {
      st->print("alias_idx==%d, adr_check==", alias_idx);
      if( adr_check == NULL ) {
        st->print("NULL");
      } else {
        adr_check->dump();
      }
      st->cr();
      print_alias_types();
      assert(consistent, "adr_check must match alias idx");
    }
  }
#endif
  // TypeInstPtr::NOTNULL+any is an OOP with unknown offset - generally
  // means an array I have not precisely typed yet.  Do not do any
  // alias stuff with it any time soon.
  const TypeOopPtr *tinst = tp->isa_oopptr();
  if( tp->base() != Type::AnyPtr &&
      !(tinst &&
        tinst->klass()->is_java_lang_Object() &&
        tinst->offset() == Type::OffsetBot) ) {
    // compress paths and change unreachable cycles to TOP
    // If not, we can update the input infinitely along a MergeMem cycle
    // Equivalent code in PhiNode::Ideal
    Node* m  = phase->transform(mmem);
    // If tranformed to a MergeMem, get the desired slice
    // Otherwise the returned node represents memory for every slice
    mem = (m->is_MergeMem())? m->as_MergeMem()->memory_at(alias_idx) : m;
    // Update input if it is progress over what we have now
  }
  return mem;
}

//--------------------------Ideal_common---------------------------------------
// Look for degenerate control and memory inputs.  Bypass MergeMem inputs.
// Unhook non-raw memories from complete (macro-expanded) initializations.
Node *MemNode::Ideal_common(PhaseGVN *phase, bool can_reshape) {
  // If our control input is a dead region, kill all below the region
  Node *ctl = in(MemNode::Control);
  if (ctl && remove_dead_region(phase, can_reshape))
    return this;

  // Ignore if memory is dead, or self-loop
  Node *mem = in(MemNode::Memory);
  if( phase->type( mem ) == Type::TOP ) return NodeSentinel; // caller will return NULL
  assert( mem != this, "dead loop in MemNode::Ideal" );

  Node *address = in(MemNode::Address);
  const Type *t_adr = phase->type( address );
  if( t_adr == Type::TOP )              return NodeSentinel; // caller will return NULL

  // Avoid independent memory operations
  Node* old_mem = mem;

  // The code which unhooks non-raw memories from complete (macro-expanded)
  // initializations was removed. After macro-expansion all stores catched
  // by Initialize node became raw stores and there is no information
  // which memory slices they modify. So it is unsafe to move any memory
  // operation above these stores. Also in most cases hooked non-raw memories
  // were already unhooked by using information from detect_ptr_independence()
  // and find_previous_store().

  if (mem->is_MergeMem()) {
    MergeMemNode* mmem = mem->as_MergeMem();
    const TypePtr *tp = t_adr->is_ptr();

    mem = step_through_mergemem(phase, mmem, tp, adr_type(), tty);
  }

  if (mem != old_mem) {
    set_req(MemNode::Memory, mem);
    return this;
  }

  // let the subclass continue analyzing...
  return NULL;
}

// Helper function for proving some simple control dominations.
// Attempt to prove that control input 'dom' dominates (or equals) 'sub'.
// Already assumes that 'dom' is available at 'sub', and that 'sub'
// is not a constant (dominated by the method's StartNode).
// Used by MemNode::find_previous_store to prove that the
// control input of a memory operation predates (dominates)
// an allocation it wants to look past.
bool MemNode::detect_dominating_control(Node* dom, Node* sub) {
  if (dom == NULL)      return false;
  if (dom->is_Proj())   dom = dom->in(0);
  if (dom->is_Start())  return true; // anything inside the method
  if (dom->is_Root())   return true; // dom 'controls' a constant
  int cnt = 20;                      // detect cycle or too much effort
  while (sub != NULL) {              // walk 'sub' up the chain to 'dom'
    if (--cnt < 0)   return false;   // in a cycle or too complex
    if (sub == dom)  return true;
    if (sub->is_Start())  return false;
    if (sub->is_Root())   return false;
    Node* up = sub->in(0);
    if (sub == up && sub->is_Region()) {
      for (uint i = 1; i < sub->req(); i++) {
        Node* in = sub->in(i);
        if (in != NULL && !in->is_top() && in != sub) {
          up = in; break;            // take any path on the way up to 'dom'
        }
      }
    }
    if (sub == up)  return false;    // some kind of tight cycle
    sub = up;
  }
  return false;
}

//---------------------detect_ptr_independence---------------------------------
// Used by MemNode::find_previous_store to prove that two base
// pointers are never equal.
// The pointers are accompanied by their associated allocations,
// if any, which have been previously discovered by the caller.
bool MemNode::detect_ptr_independence(Node* p1, AllocateNode* a1,
                                      Node* p2, AllocateNode* a2,
                                      PhaseTransform* phase) {
  // Attempt to prove that these two pointers cannot be aliased.
  // They may both manifestly be allocations, and they should differ.
  // Or, if they are not both allocations, they can be distinct constants.
  // Otherwise, one is an allocation and the other a pre-existing value.
  if (a1 == NULL && a2 == NULL) {           // neither an allocation
    return (p1 != p2) && p1->is_Con() && p2->is_Con();
  } else if (a1 != NULL && a2 != NULL) {    // both allocations
    return (a1 != a2);
  } else if (a1 != NULL) {                  // one allocation a1
    // (Note:  p2->is_Con implies p2->in(0)->is_Root, which dominates.)
    return detect_dominating_control(p2->in(0), a1->in(0));
  } else { //(a2 != NULL)                   // one allocation a2
    return detect_dominating_control(p1->in(0), a2->in(0));
  }
  return false;
}


// The logic for reordering loads and stores uses four steps:
// (a) Walk carefully past stores and initializations which we
//     can prove are independent of this load.
// (b) Observe that the next memory state makes an exact match
//     with self (load or store), and locate the relevant store.
// (c) Ensure that, if we were to wire self directly to the store,
//     the optimizer would fold it up somehow.
// (d) Do the rewiring, and return, depending on some other part of
//     the optimizer to fold up the load.
// This routine handles steps (a) and (b).  Steps (c) and (d) are
// specific to loads and stores, so they are handled by the callers.
// (Currently, only LoadNode::Ideal has steps (c), (d).  More later.)
//
Node* MemNode::find_previous_store(PhaseTransform* phase) {
  Node*         ctrl   = in(MemNode::Control);
  Node*         adr    = in(MemNode::Address);
  intptr_t      offset = 0;
  Node*         base   = AddPNode::Ideal_base_and_offset(adr, phase, offset);
  AllocateNode* alloc  = AllocateNode::Ideal_allocation(base, phase);

  if (offset == Type::OffsetBot)
    return NULL;            // cannot unalias unless there are precise offsets

  const TypeOopPtr *addr_t = adr->bottom_type()->isa_oopptr();

  intptr_t size_in_bytes = memory_size();

  Node* mem = in(MemNode::Memory);   // start searching here...

  int cnt = 50;             // Cycle limiter
  for (;;) {                // While we can dance past unrelated stores...
    if (--cnt < 0)  break;  // Caught in cycle or a complicated dance?

    if (mem->is_Store()) {
      Node* st_adr = mem->in(MemNode::Address);
      intptr_t st_offset = 0;
      Node* st_base = AddPNode::Ideal_base_and_offset(st_adr, phase, st_offset);
      if (st_base == NULL)
        break;              // inscrutable pointer
      if (st_offset != offset && st_offset != Type::OffsetBot) {
        const int MAX_STORE = BytesPerLong;
        if (st_offset >= offset + size_in_bytes ||
            st_offset <= offset - MAX_STORE ||
            st_offset <= offset - mem->as_Store()->memory_size()) {
          // Success:  The offsets are provably independent.
          // (You may ask, why not just test st_offset != offset and be done?
          // The answer is that stores of different sizes can co-exist
          // in the same sequence of RawMem effects.  We sometimes initialize
          // a whole 'tile' of array elements with a single jint or jlong.)
          mem = mem->in(MemNode::Memory);
          continue;           // (a) advance through independent store memory
        }
      }
      if (st_base != base &&
          detect_ptr_independence(base, alloc,
                                  st_base,
                                  AllocateNode::Ideal_allocation(st_base, phase),
                                  phase)) {
        // Success:  The bases are provably independent.
        mem = mem->in(MemNode::Memory);
        continue;           // (a) advance through independent store memory
      }

      // (b) At this point, if the bases or offsets do not agree, we lose,
      // since we have not managed to prove 'this' and 'mem' independent.
      if (st_base == base && st_offset == offset) {
        return mem;         // let caller handle steps (c), (d)
      }

    } else if (mem->is_Proj() && mem->in(0)->is_Initialize()) {
      InitializeNode* st_init = mem->in(0)->as_Initialize();
      AllocateNode*  st_alloc = st_init->allocation();
      if (st_alloc == NULL)
        break;              // something degenerated
      bool known_identical = false;
      bool known_independent = false;
      if (alloc == st_alloc)
        known_identical = true;
      else if (alloc != NULL)
        known_independent = true;
      else if (ctrl != NULL &&
               detect_dominating_control(ctrl, st_alloc->in(0)))
        known_independent = true;

      if (known_independent) {
        // The bases are provably independent: Either they are
        // manifestly distinct allocations, or else the control
        // of this load dominates the store's allocation.
        int alias_idx = phase->C->get_alias_index(adr_type());
        if (alias_idx == Compile::AliasIdxRaw) {
          mem = st_alloc->in(TypeFunc::Memory);
        } else {
          mem = st_init->memory(alias_idx);
        }
        continue;           // (a) advance through independent store memory
      }

      // (b) at this point, if we are not looking at a store initializing
      // the same allocation we are loading from, we lose.
      if (known_identical) {
        // From caller, can_see_stored_value will consult find_captured_store.
        return mem;         // let caller handle steps (c), (d)
      }

    } else if (addr_t != NULL && addr_t->is_instance_field()) {
      // Can't use optimize_simple_memory_chain() since it needs PhaseGVN.
      if (mem->is_Proj() && mem->in(0)->is_Call()) {
        CallNode *call = mem->in(0)->as_Call();
        if (!call->may_modify(addr_t, phase)) {
          mem = call->in(TypeFunc::Memory);
          continue;         // (a) advance through independent call memory
        }
      } else if (mem->is_Proj() && mem->in(0)->is_MemBar()) {
        mem = mem->in(0)->in(TypeFunc::Memory);
        continue;           // (a) advance through independent MemBar memory
      } else if (mem->is_MergeMem()) {
        int alias_idx = phase->C->get_alias_index(adr_type());
        mem = mem->as_MergeMem()->memory_at(alias_idx);
        continue;           // (a) advance through independent MergeMem memory
      }
    }

    // Unless there is an explicit 'continue', we must bail out here,
    // because 'mem' is an inscrutable memory state (e.g., a call).
    break;
  }

  return NULL;              // bail out
}

//----------------------calculate_adr_type-------------------------------------
// Helper function.  Notices when the given type of address hits top or bottom.
// Also, asserts a cross-check of the type against the expected address type.
const TypePtr* MemNode::calculate_adr_type(const Type* t, const TypePtr* cross_check) {
  if (t == Type::TOP)  return NULL; // does not touch memory any more?
  #ifdef PRODUCT
  cross_check = NULL;
  #else
  if (!VerifyAliases || is_error_reported() || Node::in_dump())  cross_check = NULL;
  #endif
  const TypePtr* tp = t->isa_ptr();
  if (tp == NULL) {
    assert(cross_check == NULL || cross_check == TypePtr::BOTTOM, "expected memory type must be wide");
    return TypePtr::BOTTOM;           // touches lots of memory
  } else {
    #ifdef ASSERT
    // %%%% [phh] We don't check the alias index if cross_check is
    //            TypeRawPtr::BOTTOM.  Needs to be investigated.
    if (cross_check != NULL &&
        cross_check != TypePtr::BOTTOM &&
        cross_check != TypeRawPtr::BOTTOM) {
      // Recheck the alias index, to see if it has changed (due to a bug).
      Compile* C = Compile::current();
      assert(C->get_alias_index(cross_check) == C->get_alias_index(tp),
             "must stay in the original alias category");
      // The type of the address must be contained in the adr_type,
      // disregarding "null"-ness.
      // (We make an exception for TypeRawPtr::BOTTOM, which is a bit bucket.)
      const TypePtr* tp_notnull = tp->join(TypePtr::NOTNULL)->is_ptr();
      assert(cross_check->meet(tp_notnull) == cross_check,
             "real address must not escape from expected memory type");
    }
    #endif
    return tp;
  }
}

//------------------------adr_phi_is_loop_invariant----------------------------
// A helper function for Ideal_DU_postCCP to check if a Phi in a counted
// loop is loop invariant. Make a quick traversal of Phi and associated
// CastPP nodes, looking to see if they are a closed group within the loop.
bool MemNode::adr_phi_is_loop_invariant(Node* adr_phi, Node* cast) {
  // The idea is that the phi-nest must boil down to only CastPP nodes
  // with the same data. This implies that any path into the loop already
  // includes such a CastPP, and so the original cast, whatever its input,
  // must be covered by an equivalent cast, with an earlier control input.
  ResourceMark rm;

  // The loop entry input of the phi should be the unique dominating
  // node for every Phi/CastPP in the loop.
  Unique_Node_List closure;
  closure.push(adr_phi->in(LoopNode::EntryControl));

  // Add the phi node and the cast to the worklist.
  Unique_Node_List worklist;
  worklist.push(adr_phi);
  if( cast != NULL ){
    if( !cast->is_ConstraintCast() ) return false;
    worklist.push(cast);
  }

  // Begin recursive walk of phi nodes.
  while( worklist.size() ){
    // Take a node off the worklist
    Node *n = worklist.pop();
    if( !closure.member(n) ){
      // Add it to the closure.
      closure.push(n);
      // Make a sanity check to ensure we don't waste too much time here.
      if( closure.size() > 20) return false;
      // This node is OK if:
      //  - it is a cast of an identical value
      //  - or it is a phi node (then we add its inputs to the worklist)
      // Otherwise, the node is not OK, and we presume the cast is not invariant
      if( n->is_ConstraintCast() ){
        worklist.push(n->in(1));
      } else if( n->is_Phi() ) {
        for( uint i = 1; i < n->req(); i++ ) {
          worklist.push(n->in(i));
        }
      } else {
        return false;
      }
    }
  }

  // Quit when the worklist is empty, and we've found no offending nodes.
  return true;
}

//------------------------------Ideal_DU_postCCP-------------------------------
// Find any cast-away of null-ness and keep its control.  Null cast-aways are
// going away in this pass and we need to make this memory op depend on the
// gating null check.

// I tried to leave the CastPP's in.  This makes the graph more accurate in
// some sense; we get to keep around the knowledge that an oop is not-null
// after some test.  Alas, the CastPP's interfere with GVN (some values are
// the regular oop, some are the CastPP of the oop, all merge at Phi's which
// cannot collapse, etc).  This cost us 10% on SpecJVM, even when I removed
// some of the more trivial cases in the optimizer.  Removing more useless
// Phi's started allowing Loads to illegally float above null checks.  I gave
// up on this approach.  CNC 10/20/2000
Node *MemNode::Ideal_DU_postCCP( PhaseCCP *ccp ) {
  Node *ctr = in(MemNode::Control);
  Node *mem = in(MemNode::Memory);
  Node *adr = in(MemNode::Address);
  Node *skipped_cast = NULL;
  // Need a null check?  Regular static accesses do not because they are
  // from constant addresses.  Array ops are gated by the range check (which
  // always includes a NULL check).  Just check field ops.
  if( !ctr ) {
    // Scan upwards for the highest location we can place this memory op.
    while( true ) {
      switch( adr->Opcode() ) {

      case Op_AddP:             // No change to NULL-ness, so peek thru AddP's
        adr = adr->in(AddPNode::Base);
        continue;

      case Op_DecodeN:         // No change to NULL-ness, so peek thru
        adr = adr->in(1);
        continue;

      case Op_CastPP:
        // If the CastPP is useless, just peek on through it.
        if( ccp->type(adr) == ccp->type(adr->in(1)) ) {
          // Remember the cast that we've peeked though. If we peek
          // through more than one, then we end up remembering the highest
          // one, that is, if in a loop, the one closest to the top.
          skipped_cast = adr;
          adr = adr->in(1);
          continue;
        }
        // CastPP is going away in this pass!  We need this memory op to be
        // control-dependent on the test that is guarding the CastPP.
        ccp->hash_delete(this);
        set_req(MemNode::Control, adr->in(0));
        ccp->hash_insert(this);
        return this;

      case Op_Phi:
        // Attempt to float above a Phi to some dominating point.
        if (adr->in(0) != NULL && adr->in(0)->is_CountedLoop()) {
          // If we've already peeked through a Cast (which could have set the
          // control), we can't float above a Phi, because the skipped Cast
          // may not be loop invariant.
          if (adr_phi_is_loop_invariant(adr, skipped_cast)) {
            adr = adr->in(1);
            continue;
          }
        }

        // Intentional fallthrough!

        // No obvious dominating point.  The mem op is pinned below the Phi
        // by the Phi itself.  If the Phi goes away (no true value is merged)
        // then the mem op can float, but not indefinitely.  It must be pinned
        // behind the controls leading to the Phi.
      case Op_CheckCastPP:
        // These usually stick around to change address type, however a
        // useless one can be elided and we still need to pick up a control edge
        if (adr->in(0) == NULL) {
          // This CheckCastPP node has NO control and is likely useless. But we
          // need check further up the ancestor chain for a control input to keep
          // the node in place. 4959717.
          skipped_cast = adr;
          adr = adr->in(1);
          continue;
        }
        ccp->hash_delete(this);
        set_req(MemNode::Control, adr->in(0));
        ccp->hash_insert(this);
        return this;

        // List of "safe" opcodes; those that implicitly block the memory
        // op below any null check.
      case Op_CastX2P:          // no null checks on native pointers
      case Op_Parm:             // 'this' pointer is not null
      case Op_LoadP:            // Loading from within a klass
      case Op_LoadN:            // Loading from within a klass
      case Op_LoadKlass:        // Loading from within a klass
      case Op_ConP:             // Loading from a klass
      case Op_CreateEx:         // Sucking up the guts of an exception oop
      case Op_Con:              // Reading from TLS
      case Op_CMoveP:           // CMoveP is pinned
        break;                  // No progress

      case Op_Proj:             // Direct call to an allocation routine
      case Op_SCMemProj:        // Memory state from store conditional ops
#ifdef ASSERT
        {
          assert(adr->as_Proj()->_con == TypeFunc::Parms, "must be return value");
          const Node* call = adr->in(0);
          if (call->is_CallStaticJava()) {
            const CallStaticJavaNode* call_java = call->as_CallStaticJava();
            const TypeTuple *r = call_java->tf()->range();
            assert(r->cnt() > TypeFunc::Parms, "must return value");
            const Type* ret_type = r->field_at(TypeFunc::Parms);
            assert(ret_type && ret_type->isa_ptr(), "must return pointer");
            // We further presume that this is one of
            // new_instance_Java, new_array_Java, or
            // the like, but do not assert for this.
          } else if (call->is_Allocate()) {
            // similar case to new_instance_Java, etc.
          } else if (!call->is_CallLeaf()) {
            // Projections from fetch_oop (OSR) are allowed as well.
            ShouldNotReachHere();
          }
        }
#endif
        break;
      default:
        ShouldNotReachHere();
      }
      break;
    }
  }

  return  NULL;               // No progress
}


//=============================================================================
uint LoadNode::size_of() const { return sizeof(*this); }
uint LoadNode::cmp( const Node &n ) const
{ return !Type::cmp( _type, ((LoadNode&)n)._type ); }
const Type *LoadNode::bottom_type() const { return _type; }
uint LoadNode::ideal_reg() const {
  return Matcher::base2reg[_type->base()];
}

#ifndef PRODUCT
void LoadNode::dump_spec(outputStream *st) const {
  MemNode::dump_spec(st);
  if( !Verbose && !WizardMode ) {
    // standard dump does this in Verbose and WizardMode
    st->print(" #"); _type->dump_on(st);
  }
}
#endif


//----------------------------LoadNode::make-----------------------------------
// Polymorphic factory method:
Node *LoadNode::make( PhaseGVN& gvn, Node *ctl, Node *mem, Node *adr, const TypePtr* adr_type, const Type *rt, BasicType bt ) {
  Compile* C = gvn.C;

  // sanity check the alias category against the created node type
  assert(!(adr_type->isa_oopptr() &&
           adr_type->offset() == oopDesc::klass_offset_in_bytes()),
         "use LoadKlassNode instead");
  assert(!(adr_type->isa_aryptr() &&
           adr_type->offset() == arrayOopDesc::length_offset_in_bytes()),
         "use LoadRangeNode instead");
  switch (bt) {
  case T_BOOLEAN:
  case T_BYTE:    return new (C, 3) LoadBNode(ctl, mem, adr, adr_type, rt->is_int()    );
  case T_INT:     return new (C, 3) LoadINode(ctl, mem, adr, adr_type, rt->is_int()    );
  case T_CHAR:    return new (C, 3) LoadCNode(ctl, mem, adr, adr_type, rt->is_int()    );
  case T_SHORT:   return new (C, 3) LoadSNode(ctl, mem, adr, adr_type, rt->is_int()    );
  case T_LONG:    return new (C, 3) LoadLNode(ctl, mem, adr, adr_type, rt->is_long()   );
  case T_FLOAT:   return new (C, 3) LoadFNode(ctl, mem, adr, adr_type, rt              );
  case T_DOUBLE:  return new (C, 3) LoadDNode(ctl, mem, adr, adr_type, rt              );
  case T_ADDRESS: return new (C, 3) LoadPNode(ctl, mem, adr, adr_type, rt->is_ptr()    );
  case T_OBJECT:
#ifdef _LP64
    if (adr->bottom_type()->is_narrow()) {
      const TypeNarrowOop* narrowtype;
      if (rt->isa_narrowoop()) {
        narrowtype = rt->is_narrowoop();
        rt = narrowtype->make_oopptr();
      } else {
        narrowtype = rt->is_oopptr()->make_narrowoop();
      }
      Node* load  = gvn.transform(new (C, 3) LoadNNode(ctl, mem, adr, adr_type, narrowtype));

      return new (C, 2) DecodeNNode(load, rt);
    } else
#endif
      {
        assert(!adr->bottom_type()->is_narrow(), "should have got back a narrow oop");
        return new (C, 3) LoadPNode(ctl, mem, adr, adr_type, rt->is_oopptr());
      }
  }
  ShouldNotReachHere();
  return (LoadNode*)NULL;
}

LoadLNode* LoadLNode::make_atomic(Compile *C, Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, const Type* rt) {
  bool require_atomic = true;
  return new (C, 3) LoadLNode(ctl, mem, adr, adr_type, rt->is_long(), require_atomic);
}


//------------------------------hash-------------------------------------------
uint LoadNode::hash() const {
  // unroll addition of interesting fields
  return (uintptr_t)in(Control) + (uintptr_t)in(Memory) + (uintptr_t)in(Address);
}

//---------------------------can_see_stored_value------------------------------
// This routine exists to make sure this set of tests is done the same
// everywhere.  We need to make a coordinated change: first LoadNode::Ideal
// will change the graph shape in a way which makes memory alive twice at the
// same time (uses the Oracle model of aliasing), then some
// LoadXNode::Identity will fold things back to the equivalence-class model
// of aliasing.
Node* MemNode::can_see_stored_value(Node* st, PhaseTransform* phase) const {
  Node* ld_adr = in(MemNode::Address);

  const TypeInstPtr* tp = phase->type(ld_adr)->isa_instptr();
  Compile::AliasType* atp = tp != NULL ? phase->C->alias_type(tp) : NULL;
  if (EliminateAutoBox && atp != NULL && atp->index() >= Compile::AliasIdxRaw &&
      atp->field() != NULL && !atp->field()->is_volatile()) {
    uint alias_idx = atp->index();
    bool final = atp->field()->is_final();
    Node* result = NULL;
    Node* current = st;
    // Skip through chains of MemBarNodes checking the MergeMems for
    // new states for the slice of this load.  Stop once any other
    // kind of node is encountered.  Loads from final memory can skip
    // through any kind of MemBar but normal loads shouldn't skip
    // through MemBarAcquire since the could allow them to move out of
    // a synchronized region.
    while (current->is_Proj()) {
      int opc = current->in(0)->Opcode();
      if ((final && opc == Op_MemBarAcquire) ||
          opc == Op_MemBarRelease || opc == Op_MemBarCPUOrder) {
        Node* mem = current->in(0)->in(TypeFunc::Memory);
        if (mem->is_MergeMem()) {
          MergeMemNode* merge = mem->as_MergeMem();
          Node* new_st = merge->memory_at(alias_idx);
          if (new_st == merge->base_memory()) {
            // Keep searching
            current = merge->base_memory();
            continue;
          }
          // Save the new memory state for the slice and fall through
          // to exit.
          result = new_st;
        }
      }
      break;
    }
    if (result != NULL) {
      st = result;
    }
  }


  // Loop around twice in the case Load -> Initialize -> Store.
  // (See PhaseIterGVN::add_users_to_worklist, which knows about this case.)
  for (int trip = 0; trip <= 1; trip++) {

    if (st->is_Store()) {
      Node* st_adr = st->in(MemNode::Address);
      if (!phase->eqv(st_adr, ld_adr)) {
        // Try harder before giving up...  Match raw and non-raw pointers.
        intptr_t st_off = 0;
        AllocateNode* alloc = AllocateNode::Ideal_allocation(st_adr, phase, st_off);
        if (alloc == NULL)       return NULL;
        intptr_t ld_off = 0;
        AllocateNode* allo2 = AllocateNode::Ideal_allocation(ld_adr, phase, ld_off);
        if (alloc != allo2)      return NULL;
        if (ld_off != st_off)    return NULL;
        // At this point we have proven something like this setup:
        //  A = Allocate(...)
        //  L = LoadQ(,  AddP(CastPP(, A.Parm),, #Off))
        //  S = StoreQ(, AddP(,        A.Parm  , #Off), V)
        // (Actually, we haven't yet proven the Q's are the same.)
        // In other words, we are loading from a casted version of
        // the same pointer-and-offset that we stored to.
        // Thus, we are able to replace L by V.
      }
      // Now prove that we have a LoadQ matched to a StoreQ, for some Q.
      if (store_Opcode() != st->Opcode())
        return NULL;
      return st->in(MemNode::ValueIn);
    }

    intptr_t offset = 0;  // scratch

    // A load from a freshly-created object always returns zero.
    // (This can happen after LoadNode::Ideal resets the load's memory input
    // to find_captured_store, which returned InitializeNode::zero_memory.)
    if (st->is_Proj() && st->in(0)->is_Allocate() &&
        st->in(0) == AllocateNode::Ideal_allocation(ld_adr, phase, offset) &&
        offset >= st->in(0)->as_Allocate()->minimum_header_size()) {
      // return a zero value for the load's basic type
      // (This is one of the few places where a generic PhaseTransform
      // can create new nodes.  Think of it as lazily manifesting
      // virtually pre-existing constants.)
      return phase->zerocon(memory_type());
    }

    // A load from an initialization barrier can match a captured store.
    if (st->is_Proj() && st->in(0)->is_Initialize()) {
      InitializeNode* init = st->in(0)->as_Initialize();
      AllocateNode* alloc = init->allocation();
      if (alloc != NULL &&
          alloc == AllocateNode::Ideal_allocation(ld_adr, phase, offset)) {
        // examine a captured store value
        st = init->find_captured_store(offset, memory_size(), phase);
        if (st != NULL)
          continue;             // take one more trip around
      }
    }

    break;
  }

  return NULL;
}

//----------------------is_instance_field_load_with_local_phi------------------
bool LoadNode::is_instance_field_load_with_local_phi(Node* ctrl) {
  if( in(MemNode::Memory)->is_Phi() && in(MemNode::Memory)->in(0) == ctrl &&
      in(MemNode::Address)->is_AddP() ) {
    const TypeOopPtr* t_oop = in(MemNode::Address)->bottom_type()->isa_oopptr();
    // Only instances.
    if( t_oop != NULL && t_oop->is_instance_field() &&
        t_oop->offset() != Type::OffsetBot &&
        t_oop->offset() != Type::OffsetTop) {
      return true;
    }
  }
  return false;
}

//------------------------------Identity---------------------------------------
// Loads are identity if previous store is to same address
Node *LoadNode::Identity( PhaseTransform *phase ) {
  // If the previous store-maker is the right kind of Store, and the store is
  // to the same address, then we are equal to the value stored.
  Node* mem = in(MemNode::Memory);
  Node* value = can_see_stored_value(mem, phase);
  if( value ) {
    // byte, short & char stores truncate naturally.
    // A load has to load the truncated value which requires
    // some sort of masking operation and that requires an
    // Ideal call instead of an Identity call.
    if (memory_size() < BytesPerInt) {
      // If the input to the store does not fit with the load's result type,
      // it must be truncated via an Ideal call.
      if (!phase->type(value)->higher_equal(phase->type(this)))
        return this;
    }
    // (This works even when value is a Con, but LoadNode::Value
    // usually runs first, producing the singleton type of the Con.)
    return value;
  }

  // Search for an existing data phi which was generated before for the same
  // instance's field to avoid infinite genertion of phis in a loop.
  Node *region = mem->in(0);
  if (is_instance_field_load_with_local_phi(region)) {
    const TypePtr *addr_t = in(MemNode::Address)->bottom_type()->isa_ptr();
    int this_index  = phase->C->get_alias_index(addr_t);
    int this_offset = addr_t->offset();
    int this_id    = addr_t->is_oopptr()->instance_id();
    const Type* this_type = bottom_type();
    for (DUIterator_Fast imax, i = region->fast_outs(imax); i < imax; i++) {
      Node* phi = region->fast_out(i);
      if (phi->is_Phi() && phi != mem &&
          phi->as_Phi()->is_same_inst_field(this_type, this_id, this_index, this_offset)) {
        return phi;
      }
    }
  }

  return this;
}


// Returns true if the AliasType refers to the field that holds the
// cached box array.  Currently only handles the IntegerCache case.
static bool is_autobox_cache(Compile::AliasType* atp) {
  if (atp != NULL && atp->field() != NULL) {
    ciField* field = atp->field();
    ciSymbol* klass = field->holder()->name();
    if (field->name() == ciSymbol::cache_field_name() &&
        field->holder()->uses_default_loader() &&
        klass == ciSymbol::java_lang_Integer_IntegerCache()) {
      return true;
    }
  }
  return false;
}

// Fetch the base value in the autobox array
static bool fetch_autobox_base(Compile::AliasType* atp, int& cache_offset) {
  if (atp != NULL && atp->field() != NULL) {
    ciField* field = atp->field();
    ciSymbol* klass = field->holder()->name();
    if (field->name() == ciSymbol::cache_field_name() &&
        field->holder()->uses_default_loader() &&
        klass == ciSymbol::java_lang_Integer_IntegerCache()) {
      assert(field->is_constant(), "what?");
      ciObjArray* array = field->constant_value().as_object()->as_obj_array();
      // Fetch the box object at the base of the array and get its value
      ciInstance* box = array->obj_at(0)->as_instance();
      ciInstanceKlass* ik = box->klass()->as_instance_klass();
      if (ik->nof_nonstatic_fields() == 1) {
        // This should be true nonstatic_field_at requires calling
        // nof_nonstatic_fields so check it anyway
        ciConstant c = box->field_value(ik->nonstatic_field_at(0));
        cache_offset = c.as_int();
      }
      return true;
    }
  }
  return false;
}

// Returns true if the AliasType refers to the value field of an
// autobox object.  Currently only handles Integer.
static bool is_autobox_object(Compile::AliasType* atp) {
  if (atp != NULL && atp->field() != NULL) {
    ciField* field = atp->field();
    ciSymbol* klass = field->holder()->name();
    if (field->name() == ciSymbol::value_name() &&
        field->holder()->uses_default_loader() &&
        klass == ciSymbol::java_lang_Integer()) {
      return true;
    }
  }
  return false;
}


// We're loading from an object which has autobox behaviour.
// If this object is result of a valueOf call we'll have a phi
// merging a newly allocated object and a load from the cache.
// We want to replace this load with the original incoming
// argument to the valueOf call.
Node* LoadNode::eliminate_autobox(PhaseGVN* phase) {
  Node* base = in(Address)->in(AddPNode::Base);
  if (base->is_Phi() && base->req() == 3) {
    AllocateNode* allocation = NULL;
    int allocation_index = -1;
    int load_index = -1;
    for (uint i = 1; i < base->req(); i++) {
      allocation = AllocateNode::Ideal_allocation(base->in(i), phase);
      if (allocation != NULL) {
        allocation_index = i;
        load_index = 3 - allocation_index;
        break;
      }
    }
    LoadNode* load = NULL;
    if (allocation != NULL && base->in(load_index)->is_Load()) {
      load = base->in(load_index)->as_Load();
    }
    if (load != NULL && in(Memory)->is_Phi() && in(Memory)->in(0) == base->in(0)) {
      // Push the loads from the phi that comes from valueOf up
      // through it to allow elimination of the loads and the recovery
      // of the original value.
      Node* mem_phi = in(Memory);
      Node* offset = in(Address)->in(AddPNode::Offset);

      Node* in1 = clone();
      Node* in1_addr = in1->in(Address)->clone();
      in1_addr->set_req(AddPNode::Base, base->in(allocation_index));
      in1_addr->set_req(AddPNode::Address, base->in(allocation_index));
      in1_addr->set_req(AddPNode::Offset, offset);
      in1->set_req(0, base->in(allocation_index));
      in1->set_req(Address, in1_addr);
      in1->set_req(Memory, mem_phi->in(allocation_index));

      Node* in2 = clone();
      Node* in2_addr = in2->in(Address)->clone();
      in2_addr->set_req(AddPNode::Base, base->in(load_index));
      in2_addr->set_req(AddPNode::Address, base->in(load_index));
      in2_addr->set_req(AddPNode::Offset, offset);
      in2->set_req(0, base->in(load_index));
      in2->set_req(Address, in2_addr);
      in2->set_req(Memory, mem_phi->in(load_index));

      in1_addr = phase->transform(in1_addr);
      in1 =      phase->transform(in1);
      in2_addr = phase->transform(in2_addr);
      in2 =      phase->transform(in2);

      PhiNode* result = PhiNode::make_blank(base->in(0), this);
      result->set_req(allocation_index, in1);
      result->set_req(load_index, in2);
      return result;
    }
  } else if (base->is_Load()) {
    // Eliminate the load of Integer.value for integers from the cache
    // array by deriving the value from the index into the array.
    // Capture the offset of the load and then reverse the computation.
    Node* load_base = base->in(Address)->in(AddPNode::Base);
    if (load_base != NULL) {
      Compile::AliasType* atp = phase->C->alias_type(load_base->adr_type());
      intptr_t cache_offset;
      int shift = -1;
      Node* cache = NULL;
      if (is_autobox_cache(atp)) {
        shift  = exact_log2(type2aelembytes(T_OBJECT));
        cache = AddPNode::Ideal_base_and_offset(load_base->in(Address), phase, cache_offset);
      }
      if (cache != NULL && base->in(Address)->is_AddP()) {
        Node* elements[4];
        int count = base->in(Address)->as_AddP()->unpack_offsets(elements, ARRAY_SIZE(elements));
        int cache_low;
        if (count > 0 && fetch_autobox_base(atp, cache_low)) {
          int offset = arrayOopDesc::base_offset_in_bytes(memory_type()) - (cache_low << shift);
          // Add up all the offsets making of the address of the load
          Node* result = elements[0];
          for (int i = 1; i < count; i++) {
            result = phase->transform(new (phase->C, 3) AddXNode(result, elements[i]));
          }
          // Remove the constant offset from the address and then
          // remove the scaling of the offset to recover the original index.
          result = phase->transform(new (phase->C, 3) AddXNode(result, phase->MakeConX(-offset)));
          if (result->Opcode() == Op_LShiftX && result->in(2) == phase->intcon(shift)) {
            // Peel the shift off directly but wrap it in a dummy node
            // since Ideal can't return existing nodes
            result = new (phase->C, 3) RShiftXNode(result->in(1), phase->intcon(0));
          } else {
            result = new (phase->C, 3) RShiftXNode(result, phase->intcon(shift));
          }
#ifdef _LP64
          result = new (phase->C, 2) ConvL2INode(phase->transform(result));
#endif
          return result;
        }
      }
    }
  }
  return NULL;
}


//------------------------------Ideal------------------------------------------
// If the load is from Field memory and the pointer is non-null, we can
// zero out the control input.
// If the offset is constant and the base is an object allocation,
// try to hook me up to the exact initializing store.
Node *LoadNode::Ideal(PhaseGVN *phase, bool can_reshape) {
  Node* p = MemNode::Ideal_common(phase, can_reshape);
  if (p)  return (p == NodeSentinel) ? NULL : p;

  Node* ctrl    = in(MemNode::Control);
  Node* address = in(MemNode::Address);

  // Skip up past a SafePoint control.  Cannot do this for Stores because
  // pointer stores & cardmarks must stay on the same side of a SafePoint.
  if( ctrl != NULL && ctrl->Opcode() == Op_SafePoint &&
      phase->C->get_alias_index(phase->type(address)->is_ptr()) != Compile::AliasIdxRaw ) {
    ctrl = ctrl->in(0);
    set_req(MemNode::Control,ctrl);
  }

  // Check for useless control edge in some common special cases
  if (in(MemNode::Control) != NULL) {
    intptr_t ignore = 0;
    Node*    base   = AddPNode::Ideal_base_and_offset(address, phase, ignore);
    if (base != NULL
        && phase->type(base)->higher_equal(TypePtr::NOTNULL)
        && detect_dominating_control(base->in(0), phase->C->start())) {
      // A method-invariant, non-null address (constant or 'this' argument).
      set_req(MemNode::Control, NULL);
    }
  }

  if (EliminateAutoBox && can_reshape && in(Address)->is_AddP()) {
    Node* base = in(Address)->in(AddPNode::Base);
    if (base != NULL) {
      Compile::AliasType* atp = phase->C->alias_type(adr_type());
      if (is_autobox_object(atp)) {
        Node* result = eliminate_autobox(phase);
        if (result != NULL) return result;
      }
    }
  }

  Node* mem = in(MemNode::Memory);
  const TypePtr *addr_t = phase->type(address)->isa_ptr();

  if (addr_t != NULL) {
    // try to optimize our memory input
    Node* opt_mem = MemNode::optimize_memory_chain(mem, addr_t, phase);
    if (opt_mem != mem) {
      set_req(MemNode::Memory, opt_mem);
      return this;
    }
    const TypeOopPtr *t_oop = addr_t->isa_oopptr();
    if (can_reshape && opt_mem->is_Phi() &&
        (t_oop != NULL) && t_oop->is_instance_field()) {
      assert(t_oop->offset() != Type::OffsetBot && t_oop->offset() != Type::OffsetTop, "");
      Node *region = opt_mem->in(0);
      uint cnt = opt_mem->req();
      for( uint i = 1; i < cnt; i++ ) {
        Node *in = opt_mem->in(i);
        if( in == NULL ) {
          region = NULL; // Wait stable graph
          break;
        }
      }
      if (region != NULL) {
        // Check for loop invariant.
        if (cnt == 3) {
          for( uint i = 1; i < cnt; i++ ) {
            Node *in = opt_mem->in(i);
            Node* m = MemNode::optimize_memory_chain(in, addr_t, phase);
            if (m == opt_mem) {
              set_req(MemNode::Memory, opt_mem->in(cnt - i)); // Skip this phi.
              return this;
            }
          }
        }
        // Split through Phi (see original code in loopopts.cpp).
        assert(phase->C->have_alias_type(addr_t), "instance should have alias type");

        // Do nothing here if Identity will find a value
        // (to avoid infinite chain of value phis generation).
        if ( !phase->eqv(this, this->Identity(phase)) )
          return NULL;

        const Type* this_type = this->bottom_type();
        int this_index  = phase->C->get_alias_index(addr_t);
        int this_offset = addr_t->offset();
        int this_iid    = addr_t->is_oopptr()->instance_id();
        int wins = 0;
        PhaseIterGVN *igvn = phase->is_IterGVN();
        Node *phi = new (igvn->C, region->req()) PhiNode(region, this_type, NULL, this_iid, this_index, this_offset);
        for( uint i = 1; i < region->req(); i++ ) {
          Node *x;
          Node* the_clone = NULL;
          if( region->in(i) == phase->C->top() ) {
            x = phase->C->top();      // Dead path?  Use a dead data op
          } else {
            x = this->clone();        // Else clone up the data op
            the_clone = x;            // Remember for possible deletion.
            // Alter data node to use pre-phi inputs
            if( this->in(0) == region ) {
              x->set_req( 0, region->in(i) );
            } else {
              x->set_req( 0, NULL );
            }
            for( uint j = 1; j < this->req(); j++ ) {
              Node *in = this->in(j);
              if( in->is_Phi() && in->in(0) == region )
                x->set_req( j, in->in(i) ); // Use pre-Phi input for the clone
            }
          }
          // Check for a 'win' on some paths
          const Type *t = x->Value(igvn);

          bool singleton = t->singleton();

          // See comments in PhaseIdealLoop::split_thru_phi().
          if( singleton && t == Type::TOP ) {
            singleton &= region->is_Loop() && (i != LoopNode::EntryControl);
          }

          if( singleton ) {
            wins++;
            x = igvn->makecon(t);
          } else {
            // We now call Identity to try to simplify the cloned node.
            // Note that some Identity methods call phase->type(this).
            // Make sure that the type array is big enough for
            // our new node, even though we may throw the node away.
            // (This tweaking with igvn only works because x is a new node.)
            igvn->set_type(x, t);
            Node *y = x->Identity(igvn);
            if( y != x ) {
              wins++;
              x = y;
            } else {
              y = igvn->hash_find(x);
              if( y ) {
                wins++;
                x = y;
              } else {
                // Else x is a new node we are keeping
                // We do not need register_new_node_with_optimizer
                // because set_type has already been called.
                igvn->_worklist.push(x);
              }
            }
          }
          if (x != the_clone && the_clone != NULL)
            igvn->remove_dead_node(the_clone);
          phi->set_req(i, x);
        }
        if( wins > 0 ) {
          // Record Phi
          igvn->register_new_node_with_optimizer(phi);
          return phi;
        } else {
          igvn->remove_dead_node(phi);
        }
      }
    }
  }

  // Check for prior store with a different base or offset; make Load
  // independent.  Skip through any number of them.  Bail out if the stores
  // are in an endless dead cycle and report no progress.  This is a key
  // transform for Reflection.  However, if after skipping through the Stores
  // we can't then fold up against a prior store do NOT do the transform as
  // this amounts to using the 'Oracle' model of aliasing.  It leaves the same
  // array memory alive twice: once for the hoisted Load and again after the
  // bypassed Store.  This situation only works if EVERYBODY who does
  // anti-dependence work knows how to bypass.  I.e. we need all
  // anti-dependence checks to ask the same Oracle.  Right now, that Oracle is
  // the alias index stuff.  So instead, peek through Stores and IFF we can
  // fold up, do so.
  Node* prev_mem = find_previous_store(phase);
  // Steps (a), (b):  Walk past independent stores to find an exact match.
  if (prev_mem != NULL && prev_mem != in(MemNode::Memory)) {
    // (c) See if we can fold up on the spot, but don't fold up here.
    // Fold-up might require truncation (for LoadB/LoadS/LoadC) or
    // just return a prior value, which is done by Identity calls.
    if (can_see_stored_value(prev_mem, phase)) {
      // Make ready for step (d):
      set_req(MemNode::Memory, prev_mem);
      return this;
    }
  }

  return NULL;                  // No further progress
}

// Helper to recognize certain Klass fields which are invariant across
// some group of array types (e.g., int[] or all T[] where T < Object).
const Type*
LoadNode::load_array_final_field(const TypeKlassPtr *tkls,
                                 ciKlass* klass) const {
  if (tkls->offset() == Klass::modifier_flags_offset_in_bytes() + (int)sizeof(oopDesc)) {
    // The field is Klass::_modifier_flags.  Return its (constant) value.
    // (Folds up the 2nd indirection in aClassConstant.getModifiers().)
    assert(this->Opcode() == Op_LoadI, "must load an int from _modifier_flags");
    return TypeInt::make(klass->modifier_flags());
  }
  if (tkls->offset() == Klass::access_flags_offset_in_bytes() + (int)sizeof(oopDesc)) {
    // The field is Klass::_access_flags.  Return its (constant) value.
    // (Folds up the 2nd indirection in Reflection.getClassAccessFlags(aClassConstant).)
    assert(this->Opcode() == Op_LoadI, "must load an int from _access_flags");
    return TypeInt::make(klass->access_flags());
  }
  if (tkls->offset() == Klass::layout_helper_offset_in_bytes() + (int)sizeof(oopDesc)) {
    // The field is Klass::_layout_helper.  Return its constant value if known.
    assert(this->Opcode() == Op_LoadI, "must load an int from _layout_helper");
    return TypeInt::make(klass->layout_helper());
  }

  // No match.
  return NULL;
}

//------------------------------Value-----------------------------------------
const Type *LoadNode::Value( PhaseTransform *phase ) const {
  // Either input is TOP ==> the result is TOP
  Node* mem = in(MemNode::Memory);
  const Type *t1 = phase->type(mem);
  if (t1 == Type::TOP)  return Type::TOP;
  Node* adr = in(MemNode::Address);
  const TypePtr* tp = phase->type(adr)->isa_ptr();
  if (tp == NULL || tp->empty())  return Type::TOP;
  int off = tp->offset();
  assert(off != Type::OffsetTop, "case covered by TypePtr::empty");

  // Try to guess loaded type from pointer type
  if (tp->base() == Type::AryPtr) {
    const Type *t = tp->is_aryptr()->elem();
    // Don't do this for integer types. There is only potential profit if
    // the element type t is lower than _type; that is, for int types, if _type is
    // more restrictive than t.  This only happens here if one is short and the other
    // char (both 16 bits), and in those cases we've made an intentional decision
    // to use one kind of load over the other. See AndINode::Ideal and 4965907.
    // Also, do not try to narrow the type for a LoadKlass, regardless of offset.
    //
    // Yes, it is possible to encounter an expression like (LoadKlass p1:(AddP x x 8))
    // where the _gvn.type of the AddP is wider than 8.  This occurs when an earlier
    // copy p0 of (AddP x x 8) has been proven equal to p1, and the p0 has been
    // subsumed by p1.  If p1 is on the worklist but has not yet been re-transformed,
    // it is possible that p1 will have a type like Foo*[int+]:NotNull*+any.
    // In fact, that could have been the original type of p1, and p1 could have
    // had an original form like p1:(AddP x x (LShiftL quux 3)), where the
    // expression (LShiftL quux 3) independently optimized to the constant 8.
    if ((t->isa_int() == NULL) && (t->isa_long() == NULL)
        && Opcode() != Op_LoadKlass) {
      // t might actually be lower than _type, if _type is a unique
      // concrete subclass of abstract class t.
      // Make sure the reference is not into the header, by comparing
      // the offset against the offset of the start of the array's data.
      // Different array types begin at slightly different offsets (12 vs. 16).
      // We choose T_BYTE as an example base type that is least restrictive
      // as to alignment, which will therefore produce the smallest
      // possible base offset.
      const int min_base_off = arrayOopDesc::base_offset_in_bytes(T_BYTE);
      if ((uint)off >= (uint)min_base_off) {  // is the offset beyond the header?
        const Type* jt = t->join(_type);
        // In any case, do not allow the join, per se, to empty out the type.
        if (jt->empty() && !t->empty()) {
          // This can happen if a interface-typed array narrows to a class type.
          jt = _type;
        }

        if (EliminateAutoBox) {
          // The pointers in the autobox arrays are always non-null
          Node* base = in(Address)->in(AddPNode::Base);
          if (base != NULL) {
            Compile::AliasType* atp = phase->C->alias_type(base->adr_type());
            if (is_autobox_cache(atp)) {
              return jt->join(TypePtr::NOTNULL)->is_ptr();
            }
          }
        }
        return jt;
      }
    }
  } else if (tp->base() == Type::InstPtr) {
    assert( off != Type::OffsetBot ||
            // arrays can be cast to Objects
            tp->is_oopptr()->klass()->is_java_lang_Object() ||
            // unsafe field access may not have a constant offset
            phase->C->has_unsafe_access(),
            "Field accesses must be precise" );
    // For oop loads, we expect the _type to be precise
  } else if (tp->base() == Type::KlassPtr) {
    assert( off != Type::OffsetBot ||
            // arrays can be cast to Objects
            tp->is_klassptr()->klass()->is_java_lang_Object() ||
            // also allow array-loading from the primary supertype
            // array during subtype checks
            Opcode() == Op_LoadKlass,
            "Field accesses must be precise" );
    // For klass/static loads, we expect the _type to be precise
  }

  const TypeKlassPtr *tkls = tp->isa_klassptr();
  if (tkls != NULL && !StressReflectiveCode) {
    ciKlass* klass = tkls->klass();
    if (klass->is_loaded() && tkls->klass_is_exact()) {
      // We are loading a field from a Klass metaobject whose identity
      // is known at compile time (the type is "exact" or "precise").
      // Check for fields we know are maintained as constants by the VM.
      if (tkls->offset() == Klass::super_check_offset_offset_in_bytes() + (int)sizeof(oopDesc)) {
        // The field is Klass::_super_check_offset.  Return its (constant) value.
        // (Folds up type checking code.)
        assert(Opcode() == Op_LoadI, "must load an int from _super_check_offset");
        return TypeInt::make(klass->super_check_offset());
      }
      // Compute index into primary_supers array
      juint depth = (tkls->offset() - (Klass::primary_supers_offset_in_bytes() + (int)sizeof(oopDesc))) / sizeof(klassOop);
      // Check for overflowing; use unsigned compare to handle the negative case.
      if( depth < ciKlass::primary_super_limit() ) {
        // The field is an element of Klass::_primary_supers.  Return its (constant) value.
        // (Folds up type checking code.)
        assert(Opcode() == Op_LoadKlass, "must load a klass from _primary_supers");
        ciKlass *ss = klass->super_of_depth(depth);
        return ss ? TypeKlassPtr::make(ss) : TypePtr::NULL_PTR;
      }
      const Type* aift = load_array_final_field(tkls, klass);
      if (aift != NULL)  return aift;
      if (tkls->offset() == in_bytes(arrayKlass::component_mirror_offset()) + (int)sizeof(oopDesc)
          && klass->is_array_klass()) {
        // The field is arrayKlass::_component_mirror.  Return its (constant) value.
        // (Folds up aClassConstant.getComponentType, common in Arrays.copyOf.)
        assert(Opcode() == Op_LoadP, "must load an oop from _component_mirror");
        return TypeInstPtr::make(klass->as_array_klass()->component_mirror());
      }
      if (tkls->offset() == Klass::java_mirror_offset_in_bytes() + (int)sizeof(oopDesc)) {
        // The field is Klass::_java_mirror.  Return its (constant) value.
        // (Folds up the 2nd indirection in anObjConstant.getClass().)
        assert(Opcode() == Op_LoadP, "must load an oop from _java_mirror");
        return TypeInstPtr::make(klass->java_mirror());
      }
    }

    // We can still check if we are loading from the primary_supers array at a
    // shallow enough depth.  Even though the klass is not exact, entries less
    // than or equal to its super depth are correct.
    if (klass->is_loaded() ) {
      ciType *inner = klass->klass();
      while( inner->is_obj_array_klass() )
        inner = inner->as_obj_array_klass()->base_element_type();
      if( inner->is_instance_klass() &&
          !inner->as_instance_klass()->flags().is_interface() ) {
        // Compute index into primary_supers array
        juint depth = (tkls->offset() - (Klass::primary_supers_offset_in_bytes() + (int)sizeof(oopDesc))) / sizeof(klassOop);
        // Check for overflowing; use unsigned compare to handle the negative case.
        if( depth < ciKlass::primary_super_limit() &&
            depth <= klass->super_depth() ) { // allow self-depth checks to handle self-check case
          // The field is an element of Klass::_primary_supers.  Return its (constant) value.
          // (Folds up type checking code.)
          assert(Opcode() == Op_LoadKlass, "must load a klass from _primary_supers");
          ciKlass *ss = klass->super_of_depth(depth);
          return ss ? TypeKlassPtr::make(ss) : TypePtr::NULL_PTR;
        }
      }
    }

    // If the type is enough to determine that the thing is not an array,
    // we can give the layout_helper a positive interval type.
    // This will help short-circuit some reflective code.
    if (tkls->offset() == Klass::layout_helper_offset_in_bytes() + (int)sizeof(oopDesc)
        && !klass->is_array_klass() // not directly typed as an array
        && !klass->is_interface()  // specifically not Serializable & Cloneable
        && !klass->is_java_lang_Object()   // not the supertype of all T[]
        ) {
      // Note:  When interfaces are reliable, we can narrow the interface
      // test to (klass != Serializable && klass != Cloneable).
      assert(Opcode() == Op_LoadI, "must load an int from _layout_helper");
      jint min_size = Klass::instance_layout_helper(oopDesc::header_size(), false);
      // The key property of this type is that it folds up tests
      // for array-ness, since it proves that the layout_helper is positive.
      // Thus, a generic value like the basic object layout helper works fine.
      return TypeInt::make(min_size, max_jint, Type::WidenMin);
    }
  }

  // If we are loading from a freshly-allocated object, produce a zero,
  // if the load is provably beyond the header of the object.
  // (Also allow a variable load from a fresh array to produce zero.)
  if (ReduceFieldZeroing) {
    Node* value = can_see_stored_value(mem,phase);
    if (value != NULL && value->is_Con())
      return value->bottom_type();
  }

  const TypeOopPtr *tinst = tp->isa_oopptr();
  if (tinst != NULL && tinst->is_instance_field()) {
    // If we have an instance type and our memory input is the
    // programs's initial memory state, there is no matching store,
    // so just return a zero of the appropriate type
    Node *mem = in(MemNode::Memory);
    if (mem->is_Parm() && mem->in(0)->is_Start()) {
      assert(mem->as_Parm()->_con == TypeFunc::Memory, "must be memory Parm");
      return Type::get_zero_type(_type->basic_type());
    }
  }
  return _type;
}

//------------------------------match_edge-------------------------------------
// Do we Match on this edge index or not?  Match only the address.
uint LoadNode::match_edge(uint idx) const {
  return idx == MemNode::Address;
}

//--------------------------LoadBNode::Ideal--------------------------------------
//
//  If the previous store is to the same address as this load,
//  and the value stored was larger than a byte, replace this load
//  with the value stored truncated to a byte.  If no truncation is
//  needed, the replacement is done in LoadNode::Identity().
//
Node *LoadBNode::Ideal(PhaseGVN *phase, bool can_reshape) {
  Node* mem = in(MemNode::Memory);
  Node* value = can_see_stored_value(mem,phase);
  if( value && !phase->type(value)->higher_equal( _type ) ) {
    Node *result = phase->transform( new (phase->C, 3) LShiftINode(value, phase->intcon(24)) );
    return new (phase->C, 3) RShiftINode(result, phase->intcon(24));
  }
  // Identity call will handle the case where truncation is not needed.
  return LoadNode::Ideal(phase, can_reshape);
}

//--------------------------LoadCNode::Ideal--------------------------------------
//
//  If the previous store is to the same address as this load,
//  and the value stored was larger than a char, replace this load
//  with the value stored truncated to a char.  If no truncation is
//  needed, the replacement is done in LoadNode::Identity().
//
Node *LoadCNode::Ideal(PhaseGVN *phase, bool can_reshape) {
  Node* mem = in(MemNode::Memory);
  Node* value = can_see_stored_value(mem,phase);
  if( value && !phase->type(value)->higher_equal( _type ) )
    return new (phase->C, 3) AndINode(value,phase->intcon(0xFFFF));
  // Identity call will handle the case where truncation is not needed.
  return LoadNode::Ideal(phase, can_reshape);
}

//--------------------------LoadSNode::Ideal--------------------------------------
//
//  If the previous store is to the same address as this load,
//  and the value stored was larger than a short, replace this load
//  with the value stored truncated to a short.  If no truncation is
//  needed, the replacement is done in LoadNode::Identity().
//
Node *LoadSNode::Ideal(PhaseGVN *phase, bool can_reshape) {
  Node* mem = in(MemNode::Memory);
  Node* value = can_see_stored_value(mem,phase);
  if( value && !phase->type(value)->higher_equal( _type ) ) {
    Node *result = phase->transform( new (phase->C, 3) LShiftINode(value, phase->intcon(16)) );
    return new (phase->C, 3) RShiftINode(result, phase->intcon(16));
  }
  // Identity call will handle the case where truncation is not needed.
  return LoadNode::Ideal(phase, can_reshape);
}

//=============================================================================
//------------------------------Value------------------------------------------
const Type *LoadKlassNode::Value( PhaseTransform *phase ) const {
  // Either input is TOP ==> the result is TOP
  const Type *t1 = phase->type( in(MemNode::Memory) );
  if (t1 == Type::TOP)  return Type::TOP;
  Node *adr = in(MemNode::Address);
  const Type *t2 = phase->type( adr );
  if (t2 == Type::TOP)  return Type::TOP;
  const TypePtr *tp = t2->is_ptr();
  if (TypePtr::above_centerline(tp->ptr()) ||
      tp->ptr() == TypePtr::Null)  return Type::TOP;

  // Return a more precise klass, if possible
  const TypeInstPtr *tinst = tp->isa_instptr();
  if (tinst != NULL) {
    ciInstanceKlass* ik = tinst->klass()->as_instance_klass();
    int offset = tinst->offset();
    if (ik == phase->C->env()->Class_klass()
        && (offset == java_lang_Class::klass_offset_in_bytes() ||
            offset == java_lang_Class::array_klass_offset_in_bytes())) {
      // We are loading a special hidden field from a Class mirror object,
      // the field which points to the VM's Klass metaobject.
      ciType* t = tinst->java_mirror_type();
      // java_mirror_type returns non-null for compile-time Class constants.
      if (t != NULL) {
        // constant oop => constant klass
        if (offset == java_lang_Class::array_klass_offset_in_bytes()) {
          return TypeKlassPtr::make(ciArrayKlass::make(t));
        }
        if (!t->is_klass()) {
          // a primitive Class (e.g., int.class) has NULL for a klass field
          return TypePtr::NULL_PTR;
        }
        // (Folds up the 1st indirection in aClassConstant.getModifiers().)
        return TypeKlassPtr::make(t->as_klass());
      }
      // non-constant mirror, so we can't tell what's going on
    }
    if( !ik->is_loaded() )
      return _type;             // Bail out if not loaded
    if (offset == oopDesc::klass_offset_in_bytes()) {
      if (tinst->klass_is_exact()) {
        return TypeKlassPtr::make(ik);
      }
      // See if we can become precise: no subklasses and no interface
      // (Note:  We need to support verified interfaces.)
      if (!ik->is_interface() && !ik->has_subklass()) {
        //assert(!UseExactTypes, "this code should be useless with exact types");
        // Add a dependence; if any subclass added we need to recompile
        if (!ik->is_final()) {
          // %%% should use stronger assert_unique_concrete_subtype instead
          phase->C->dependencies()->assert_leaf_type(ik);
        }
        // Return precise klass
        return TypeKlassPtr::make(ik);
      }

      // Return root of possible klass
      return TypeKlassPtr::make(TypePtr::NotNull, ik, 0/*offset*/);
    }
  }

  // Check for loading klass from an array
  const TypeAryPtr *tary = tp->isa_aryptr();
  if( tary != NULL ) {
    ciKlass *tary_klass = tary->klass();
    if (tary_klass != NULL   // can be NULL when at BOTTOM or TOP
        && tary->offset() == oopDesc::klass_offset_in_bytes()) {
      if (tary->klass_is_exact()) {
        return TypeKlassPtr::make(tary_klass);
      }
      ciArrayKlass *ak = tary->klass()->as_array_klass();
      // If the klass is an object array, we defer the question to the
      // array component klass.
      if( ak->is_obj_array_klass() ) {
        assert( ak->is_loaded(), "" );
        ciKlass *base_k = ak->as_obj_array_klass()->base_element_klass();
        if( base_k->is_loaded() && base_k->is_instance_klass() ) {
          ciInstanceKlass* ik = base_k->as_instance_klass();
          // See if we can become precise: no subklasses and no interface
          if (!ik->is_interface() && !ik->has_subklass()) {
            //assert(!UseExactTypes, "this code should be useless with exact types");
            // Add a dependence; if any subclass added we need to recompile
            if (!ik->is_final()) {
              phase->C->dependencies()->assert_leaf_type(ik);
            }
            // Return precise array klass
            return TypeKlassPtr::make(ak);
          }
        }
        return TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
      } else {                  // Found a type-array?
        //assert(!UseExactTypes, "this code should be useless with exact types");
        assert( ak->is_type_array_klass(), "" );
        return TypeKlassPtr::make(ak); // These are always precise
      }
    }
  }

  // Check for loading klass from an array klass
  const TypeKlassPtr *tkls = tp->isa_klassptr();
  if (tkls != NULL && !StressReflectiveCode) {
    ciKlass* klass = tkls->klass();
    if( !klass->is_loaded() )
      return _type;             // Bail out if not loaded
    if( klass->is_obj_array_klass() &&
        (uint)tkls->offset() == objArrayKlass::element_klass_offset_in_bytes() + sizeof(oopDesc)) {
      ciKlass* elem = klass->as_obj_array_klass()->element_klass();
      // // Always returning precise element type is incorrect,
      // // e.g., element type could be object and array may contain strings
      // return TypeKlassPtr::make(TypePtr::Constant, elem, 0);

      // The array's TypeKlassPtr was declared 'precise' or 'not precise'
      // according to the element type's subclassing.
      return TypeKlassPtr::make(tkls->ptr(), elem, 0/*offset*/);
    }
    if( klass->is_instance_klass() && tkls->klass_is_exact() &&
        (uint)tkls->offset() == Klass::super_offset_in_bytes() + sizeof(oopDesc)) {
      ciKlass* sup = klass->as_instance_klass()->super();
      // The field is Klass::_super.  Return its (constant) value.
      // (Folds up the 2nd indirection in aClassConstant.getSuperClass().)
      return sup ? TypeKlassPtr::make(sup) : TypePtr::NULL_PTR;
    }
  }

  // Bailout case
  return LoadNode::Value(phase);
}

//------------------------------Identity---------------------------------------
// To clean up reflective code, simplify k.java_mirror.as_klass to plain k.
// Also feed through the klass in Allocate(...klass...)._klass.
Node* LoadKlassNode::Identity( PhaseTransform *phase ) {
  Node* x = LoadNode::Identity(phase);
  if (x != this)  return x;

  // Take apart the address into an oop and and offset.
  // Return 'this' if we cannot.
  Node*    adr    = in(MemNode::Address);
  intptr_t offset = 0;
  Node*    base   = AddPNode::Ideal_base_and_offset(adr, phase, offset);
  if (base == NULL)     return this;
  const TypeOopPtr* toop = phase->type(adr)->isa_oopptr();
  if (toop == NULL)     return this;

  // We can fetch the klass directly through an AllocateNode.
  // This works even if the klass is not constant (clone or newArray).
  if (offset == oopDesc::klass_offset_in_bytes()) {
    Node* allocated_klass = AllocateNode::Ideal_klass(base, phase);
    if (allocated_klass != NULL) {
      return allocated_klass;
    }
  }

  // Simplify k.java_mirror.as_klass to plain k, where k is a klassOop.
  // Simplify ak.component_mirror.array_klass to plain ak, ak an arrayKlass.
  // See inline_native_Class_query for occurrences of these patterns.
  // Java Example:  x.getClass().isAssignableFrom(y)
  // Java Example:  Array.newInstance(x.getClass().getComponentType(), n)
  //
  // This improves reflective code, often making the Class
  // mirror go completely dead.  (Current exception:  Class
  // mirrors may appear in debug info, but we could clean them out by
  // introducing a new debug info operator for klassOop.java_mirror).
  if (toop->isa_instptr() && toop->klass() == phase->C->env()->Class_klass()
      && (offset == java_lang_Class::klass_offset_in_bytes() ||
          offset == java_lang_Class::array_klass_offset_in_bytes())) {
    // We are loading a special hidden field from a Class mirror,
    // the field which points to its Klass or arrayKlass metaobject.
    if (base->is_Load()) {
      Node* adr2 = base->in(MemNode::Address);
      const TypeKlassPtr* tkls = phase->type(adr2)->isa_klassptr();
      if (tkls != NULL && !tkls->empty()
          && (tkls->klass()->is_instance_klass() ||
              tkls->klass()->is_array_klass())
          && adr2->is_AddP()
          ) {
        int mirror_field = Klass::java_mirror_offset_in_bytes();
        if (offset == java_lang_Class::array_klass_offset_in_bytes()) {
          mirror_field = in_bytes(arrayKlass::component_mirror_offset());
        }
        if (tkls->offset() == mirror_field + (int)sizeof(oopDesc)) {
          return adr2->in(AddPNode::Base);
        }
      }
    }
  }

  return this;
}

//------------------------------Value-----------------------------------------
const Type *LoadRangeNode::Value( PhaseTransform *phase ) const {
  // Either input is TOP ==> the result is TOP
  const Type *t1 = phase->type( in(MemNode::Memory) );
  if( t1 == Type::TOP ) return Type::TOP;
  Node *adr = in(MemNode::Address);
  const Type *t2 = phase->type( adr );
  if( t2 == Type::TOP ) return Type::TOP;
  const TypePtr *tp = t2->is_ptr();
  if (TypePtr::above_centerline(tp->ptr()))  return Type::TOP;
  const TypeAryPtr *tap = tp->isa_aryptr();
  if( !tap ) return _type;
  return tap->size();
}

//------------------------------Identity---------------------------------------
// Feed through the length in AllocateArray(...length...)._length.
Node* LoadRangeNode::Identity( PhaseTransform *phase ) {
  Node* x = LoadINode::Identity(phase);
  if (x != this)  return x;

  // Take apart the address into an oop and and offset.
  // Return 'this' if we cannot.
  Node*    adr    = in(MemNode::Address);
  intptr_t offset = 0;
  Node*    base   = AddPNode::Ideal_base_and_offset(adr, phase, offset);
  if (base == NULL)     return this;
  const TypeAryPtr* tary = phase->type(adr)->isa_aryptr();
  if (tary == NULL)     return this;

  // We can fetch the length directly through an AllocateArrayNode.
  // This works even if the length is not constant (clone or newArray).
  if (offset == arrayOopDesc::length_offset_in_bytes()) {
    Node* allocated_length = AllocateArrayNode::Ideal_length(base, phase);
    if (allocated_length != NULL) {
      return allocated_length;
    }
  }

  return this;

}
//=============================================================================
//---------------------------StoreNode::make-----------------------------------
// Polymorphic factory method:
StoreNode* StoreNode::make( PhaseGVN& gvn, Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, Node* val, BasicType bt ) {
  Compile* C = gvn.C;

  switch (bt) {
  case T_BOOLEAN:
  case T_BYTE:    return new (C, 4) StoreBNode(ctl, mem, adr, adr_type, val);
  case T_INT:     return new (C, 4) StoreINode(ctl, mem, adr, adr_type, val);
  case T_CHAR:
  case T_SHORT:   return new (C, 4) StoreCNode(ctl, mem, adr, adr_type, val);
  case T_LONG:    return new (C, 4) StoreLNode(ctl, mem, adr, adr_type, val);
  case T_FLOAT:   return new (C, 4) StoreFNode(ctl, mem, adr, adr_type, val);
  case T_DOUBLE:  return new (C, 4) StoreDNode(ctl, mem, adr, adr_type, val);
  case T_ADDRESS:
  case T_OBJECT:
#ifdef _LP64
    if (adr->bottom_type()->is_narrow() ||
        (UseCompressedOops && val->bottom_type()->isa_klassptr() &&
         adr->bottom_type()->isa_rawptr())) {
      const TypePtr* type = val->bottom_type()->is_ptr();
      Node* cp;
      if (type->isa_oopptr()) {
        const TypeNarrowOop* etype = type->is_oopptr()->make_narrowoop();
        cp = gvn.transform(new (C, 2) EncodePNode(val, etype));
      } else if (type == TypePtr::NULL_PTR) {
        cp = gvn.transform(new (C, 1) ConNNode(TypeNarrowOop::NULL_PTR));
      } else {
        ShouldNotReachHere();
      }
      return new (C, 4) StoreNNode(ctl, mem, adr, adr_type, cp);
    } else
#endif
      {
        return new (C, 4) StorePNode(ctl, mem, adr, adr_type, val);
      }
  }
  ShouldNotReachHere();
  return (StoreNode*)NULL;
}

StoreLNode* StoreLNode::make_atomic(Compile *C, Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, Node* val) {
  bool require_atomic = true;
  return new (C, 4) StoreLNode(ctl, mem, adr, adr_type, val, require_atomic);
}


//--------------------------bottom_type----------------------------------------
const Type *StoreNode::bottom_type() const {
  return Type::MEMORY;
}

//------------------------------hash-------------------------------------------
uint StoreNode::hash() const {
  // unroll addition of interesting fields
  //return (uintptr_t)in(Control) + (uintptr_t)in(Memory) + (uintptr_t)in(Address) + (uintptr_t)in(ValueIn);

  // Since they are not commoned, do not hash them:
  return NO_HASH;
}

//------------------------------Ideal------------------------------------------
// Change back-to-back Store(, p, x) -> Store(m, p, y) to Store(m, p, x).
// When a store immediately follows a relevant allocation/initialization,
// try to capture it into the initialization, or hoist it above.
Node *StoreNode::Ideal(PhaseGVN *phase, bool can_reshape) {
  Node* p = MemNode::Ideal_common(phase, can_reshape);
  if (p)  return (p == NodeSentinel) ? NULL : p;

  Node* mem     = in(MemNode::Memory);
  Node* address = in(MemNode::Address);

  // Back-to-back stores to same address?  Fold em up.
  // Generally unsafe if I have intervening uses...
  if (mem->is_Store() && phase->eqv_uncast(mem->in(MemNode::Address), address)) {
    // Looking at a dead closed cycle of memory?
    assert(mem != mem->in(MemNode::Memory), "dead loop in StoreNode::Ideal");

    assert(Opcode() == mem->Opcode() ||
           phase->C->get_alias_index(adr_type()) == Compile::AliasIdxRaw,
           "no mismatched stores, except on raw memory");

    if (mem->outcnt() == 1 &&           // check for intervening uses
        mem->as_Store()->memory_size() <= this->memory_size()) {
      // If anybody other than 'this' uses 'mem', we cannot fold 'mem' away.
      // For example, 'mem' might be the final state at a conditional return.
      // Or, 'mem' might be used by some node which is live at the same time
      // 'this' is live, which might be unschedulable.  So, require exactly
      // ONE user, the 'this' store, until such time as we clone 'mem' for
      // each of 'mem's uses (thus making the exactly-1-user-rule hold true).
      if (can_reshape) {  // (%%% is this an anachronism?)
        set_req_X(MemNode::Memory, mem->in(MemNode::Memory),
                  phase->is_IterGVN());
      } else {
        // It's OK to do this in the parser, since DU info is always accurate,
        // and the parser always refers to nodes via SafePointNode maps.
        set_req(MemNode::Memory, mem->in(MemNode::Memory));
      }
      return this;
    }
  }

  // Capture an unaliased, unconditional, simple store into an initializer.
  // Or, if it is independent of the allocation, hoist it above the allocation.
  if (ReduceFieldZeroing && /*can_reshape &&*/
      mem->is_Proj() && mem->in(0)->is_Initialize()) {
    InitializeNode* init = mem->in(0)->as_Initialize();
    intptr_t offset = init->can_capture_store(this, phase);
    if (offset > 0) {
      Node* moved = init->capture_store(this, offset, phase);
      // If the InitializeNode captured me, it made a raw copy of me,
      // and I need to disappear.
      if (moved != NULL) {
        // %%% hack to ensure that Ideal returns a new node:
        mem = MergeMemNode::make(phase->C, mem);
        return mem;             // fold me away
      }
    }
  }

  return NULL;                  // No further progress
}

//------------------------------Value-----------------------------------------
const Type *StoreNode::Value( PhaseTransform *phase ) const {
  // Either input is TOP ==> the result is TOP
  const Type *t1 = phase->type( in(MemNode::Memory) );
  if( t1 == Type::TOP ) return Type::TOP;
  const Type *t2 = phase->type( in(MemNode::Address) );
  if( t2 == Type::TOP ) return Type::TOP;
  const Type *t3 = phase->type( in(MemNode::ValueIn) );
  if( t3 == Type::TOP ) return Type::TOP;
  return Type::MEMORY;
}

//------------------------------Identity---------------------------------------
// Remove redundant stores:
//   Store(m, p, Load(m, p)) changes to m.
//   Store(, p, x) -> Store(m, p, x) changes to Store(m, p, x).
Node *StoreNode::Identity( PhaseTransform *phase ) {
  Node* mem = in(MemNode::Memory);
  Node* adr = in(MemNode::Address);
  Node* val = in(MemNode::ValueIn);

  // Load then Store?  Then the Store is useless
  if (val->is_Load() &&
      phase->eqv_uncast( val->in(MemNode::Address), adr ) &&
      phase->eqv_uncast( val->in(MemNode::Memory ), mem ) &&
      val->as_Load()->store_Opcode() == Opcode()) {
    return mem;
  }

  // Two stores in a row of the same value?
  if (mem->is_Store() &&
      phase->eqv_uncast( mem->in(MemNode::Address), adr ) &&
      phase->eqv_uncast( mem->in(MemNode::ValueIn), val ) &&
      mem->Opcode() == Opcode()) {
    return mem;
  }

  // Store of zero anywhere into a freshly-allocated object?
  // Then the store is useless.
  // (It must already have been captured by the InitializeNode.)
  if (ReduceFieldZeroing && phase->type(val)->is_zero_type()) {
    // a newly allocated object is already all-zeroes everywhere
    if (mem->is_Proj() && mem->in(0)->is_Allocate()) {
      return mem;
    }

    // the store may also apply to zero-bits in an earlier object
    Node* prev_mem = find_previous_store(phase);
    // Steps (a), (b):  Walk past independent stores to find an exact match.
    if (prev_mem != NULL) {
      Node* prev_val = can_see_stored_value(prev_mem, phase);
      if (prev_val != NULL && phase->eqv(prev_val, val)) {
        // prev_val and val might differ by a cast; it would be good
        // to keep the more informative of the two.
        return mem;
      }
    }
  }

  return this;
}

//------------------------------match_edge-------------------------------------
// Do we Match on this edge index or not?  Match only memory & value
uint StoreNode::match_edge(uint idx) const {
  return idx == MemNode::Address || idx == MemNode::ValueIn;
}

//------------------------------cmp--------------------------------------------
// Do not common stores up together.  They generally have to be split
// back up anyways, so do not bother.
uint StoreNode::cmp( const Node &n ) const {
  return (&n == this);          // Always fail except on self
}

//------------------------------Ideal_masked_input-----------------------------
// Check for a useless mask before a partial-word store
// (StoreB ... (AndI valIn conIa) )
// If (conIa & mask == mask) this simplifies to
// (StoreB ... (valIn) )
Node *StoreNode::Ideal_masked_input(PhaseGVN *phase, uint mask) {
  Node *val = in(MemNode::ValueIn);
  if( val->Opcode() == Op_AndI ) {
    const TypeInt *t = phase->type( val->in(2) )->isa_int();
    if( t && t->is_con() && (t->get_con() & mask) == mask ) {
      set_req(MemNode::ValueIn, val->in(1));
      return this;
    }
  }
  return NULL;
}


//------------------------------Ideal_sign_extended_input----------------------
// Check for useless sign-extension before a partial-word store
// (StoreB ... (RShiftI _ (LShiftI _ valIn conIL ) conIR) )
// If (conIL == conIR && conIR <= num_bits)  this simplifies to
// (StoreB ... (valIn) )
Node *StoreNode::Ideal_sign_extended_input(PhaseGVN *phase, int num_bits) {
  Node *val = in(MemNode::ValueIn);
  if( val->Opcode() == Op_RShiftI ) {
    const TypeInt *t = phase->type( val->in(2) )->isa_int();
    if( t && t->is_con() && (t->get_con() <= num_bits) ) {
      Node *shl = val->in(1);
      if( shl->Opcode() == Op_LShiftI ) {
        const TypeInt *t2 = phase->type( shl->in(2) )->isa_int();
        if( t2 && t2->is_con() && (t2->get_con() == t->get_con()) ) {
          set_req(MemNode::ValueIn, shl->in(1));
          return this;
        }
      }
    }
  }
  return NULL;
}

//------------------------------value_never_loaded-----------------------------------
// Determine whether there are any possible loads of the value stored.
// For simplicity, we actually check if there are any loads from the
// address stored to, not just for loads of the value stored by this node.
//
bool StoreNode::value_never_loaded( PhaseTransform *phase) const {
  Node *adr = in(Address);
  const TypeOopPtr *adr_oop = phase->type(adr)->isa_oopptr();
  if (adr_oop == NULL)
    return false;
  if (!adr_oop->is_instance_field())
    return false; // if not a distinct instance, there may be aliases of the address
  for (DUIterator_Fast imax, i = adr->fast_outs(imax); i < imax; i++) {
    Node *use = adr->fast_out(i);
    int opc = use->Opcode();
    if (use->is_Load() || use->is_LoadStore()) {
      return false;
    }
  }
  return true;
}

//=============================================================================
//------------------------------Ideal------------------------------------------
// If the store is from an AND mask that leaves the low bits untouched, then
// we can skip the AND operation.  If the store is from a sign-extension
// (a left shift, then right shift) we can skip both.
Node *StoreBNode::Ideal(PhaseGVN *phase, bool can_reshape){
  Node *progress = StoreNode::Ideal_masked_input(phase, 0xFF);
  if( progress != NULL ) return progress;

  progress = StoreNode::Ideal_sign_extended_input(phase, 24);
  if( progress != NULL ) return progress;

  // Finally check the default case
  return StoreNode::Ideal(phase, can_reshape);
}

//=============================================================================
//------------------------------Ideal------------------------------------------
// If the store is from an AND mask that leaves the low bits untouched, then
// we can skip the AND operation
Node *StoreCNode::Ideal(PhaseGVN *phase, bool can_reshape){
  Node *progress = StoreNode::Ideal_masked_input(phase, 0xFFFF);
  if( progress != NULL ) return progress;

  progress = StoreNode::Ideal_sign_extended_input(phase, 16);
  if( progress != NULL ) return progress;

  // Finally check the default case
  return StoreNode::Ideal(phase, can_reshape);
}

//=============================================================================
//------------------------------Identity---------------------------------------
Node *StoreCMNode::Identity( PhaseTransform *phase ) {
  // No need to card mark when storing a null ptr
  Node* my_store = in(MemNode::OopStore);
  if (my_store->is_Store()) {
    const Type *t1 = phase->type( my_store->in(MemNode::ValueIn) );
    if( t1 == TypePtr::NULL_PTR ) {
      return in(MemNode::Memory);
    }
  }
  return this;
}

//------------------------------Value-----------------------------------------
const Type *StoreCMNode::Value( PhaseTransform *phase ) const {
  // Either input is TOP ==> the result is TOP
  const Type *t = phase->type( in(MemNode::Memory) );
  if( t == Type::TOP ) return Type::TOP;
  t = phase->type( in(MemNode::Address) );
  if( t == Type::TOP ) return Type::TOP;
  t = phase->type( in(MemNode::ValueIn) );
  if( t == Type::TOP ) return Type::TOP;
  // If extra input is TOP ==> the result is TOP
  t = phase->type( in(MemNode::OopStore) );
  if( t == Type::TOP ) return Type::TOP;

  return StoreNode::Value( phase );
}


//=============================================================================
//----------------------------------SCMemProjNode------------------------------
const Type * SCMemProjNode::Value( PhaseTransform *phase ) const
{
  return bottom_type();
}

//=============================================================================
LoadStoreNode::LoadStoreNode( Node *c, Node *mem, Node *adr, Node *val, Node *ex ) : Node(5) {
  init_req(MemNode::Control, c  );
  init_req(MemNode::Memory , mem);
  init_req(MemNode::Address, adr);
  init_req(MemNode::ValueIn, val);
  init_req(         ExpectedIn, ex );
  init_class_id(Class_LoadStore);

}

//=============================================================================
//-------------------------------adr_type--------------------------------------
// Do we Match on this edge index or not?  Do not match memory
const TypePtr* ClearArrayNode::adr_type() const {
  Node *adr = in(3);
  return MemNode::calculate_adr_type(adr->bottom_type());
}

//------------------------------match_edge-------------------------------------
// Do we Match on this edge index or not?  Do not match memory
uint ClearArrayNode::match_edge(uint idx) const {
  return idx > 1;
}

//------------------------------Identity---------------------------------------
// Clearing a zero length array does nothing
Node *ClearArrayNode::Identity( PhaseTransform *phase ) {
  return phase->type(in(2))->higher_equal(TypeX::ZERO)  ? in(1) : this;
}

//------------------------------Idealize---------------------------------------
// Clearing a short array is faster with stores
Node *ClearArrayNode::Ideal(PhaseGVN *phase, bool can_reshape){
  const int unit = BytesPerLong;
  const TypeX* t = phase->type(in(2))->isa_intptr_t();
  if (!t)  return NULL;
  if (!t->is_con())  return NULL;
  intptr_t raw_count = t->get_con();
  intptr_t size = raw_count;
  if (!Matcher::init_array_count_is_in_bytes) size *= unit;
  // Clearing nothing uses the Identity call.
  // Negative clears are possible on dead ClearArrays
  // (see jck test stmt114.stmt11402.val).
  if (size <= 0 || size % unit != 0)  return NULL;
  intptr_t count = size / unit;
  // Length too long; use fast hardware clear
  if (size > Matcher::init_array_short_size)  return NULL;
  Node *mem = in(1);
  if( phase->type(mem)==Type::TOP ) return NULL;
  Node *adr = in(3);
  const Type* at = phase->type(adr);
  if( at==Type::TOP ) return NULL;
  const TypePtr* atp = at->isa_ptr();
  // adjust atp to be the correct array element address type
  if (atp == NULL)  atp = TypePtr::BOTTOM;
  else              atp = atp->add_offset(Type::OffsetBot);
  // Get base for derived pointer purposes
  if( adr->Opcode() != Op_AddP ) Unimplemented();
  Node *base = adr->in(1);

  Node *zero = phase->makecon(TypeLong::ZERO);
  Node *off  = phase->MakeConX(BytesPerLong);
  mem = new (phase->C, 4) StoreLNode(in(0),mem,adr,atp,zero);
  count--;
  while( count-- ) {
    mem = phase->transform(mem);
    adr = phase->transform(new (phase->C, 4) AddPNode(base,adr,off));
    mem = new (phase->C, 4) StoreLNode(in(0),mem,adr,atp,zero);
  }
  return mem;
}

//----------------------------clear_memory-------------------------------------
// Generate code to initialize object storage to zero.
Node* ClearArrayNode::clear_memory(Node* ctl, Node* mem, Node* dest,
                                   intptr_t start_offset,
                                   Node* end_offset,
                                   PhaseGVN* phase) {
  Compile* C = phase->C;
  intptr_t offset = start_offset;

  int unit = BytesPerLong;
  if ((offset % unit) != 0) {
    Node* adr = new (C, 4) AddPNode(dest, dest, phase->MakeConX(offset));
    adr = phase->transform(adr);
    const TypePtr* atp = TypeRawPtr::BOTTOM;
    mem = StoreNode::make(*phase, ctl, mem, adr, atp, phase->zerocon(T_INT), T_INT);
    mem = phase->transform(mem);
    offset += BytesPerInt;
  }
  assert((offset % unit) == 0, "");

  // Initialize the remaining stuff, if any, with a ClearArray.
  return clear_memory(ctl, mem, dest, phase->MakeConX(offset), end_offset, phase);
}

Node* ClearArrayNode::clear_memory(Node* ctl, Node* mem, Node* dest,
                                   Node* start_offset,
                                   Node* end_offset,
                                   PhaseGVN* phase) {
  if (start_offset == end_offset) {
    // nothing to do
    return mem;
  }

  Compile* C = phase->C;
  int unit = BytesPerLong;
  Node* zbase = start_offset;
  Node* zend  = end_offset;

  // Scale to the unit required by the CPU:
  if (!Matcher::init_array_count_is_in_bytes) {
    Node* shift = phase->intcon(exact_log2(unit));
    zbase = phase->transform( new(C,3) URShiftXNode(zbase, shift) );
    zend  = phase->transform( new(C,3) URShiftXNode(zend,  shift) );
  }

  Node* zsize = phase->transform( new(C,3) SubXNode(zend, zbase) );
  Node* zinit = phase->zerocon((unit == BytesPerLong) ? T_LONG : T_INT);

  // Bulk clear double-words
  Node* adr = phase->transform( new(C,4) AddPNode(dest, dest, start_offset) );
  mem = new (C, 4) ClearArrayNode(ctl, mem, zsize, adr);
  return phase->transform(mem);
}

Node* ClearArrayNode::clear_memory(Node* ctl, Node* mem, Node* dest,
                                   intptr_t start_offset,
                                   intptr_t end_offset,
                                   PhaseGVN* phase) {
  if (start_offset == end_offset) {
    // nothing to do
    return mem;
  }

  Compile* C = phase->C;
  assert((end_offset % BytesPerInt) == 0, "odd end offset");
  intptr_t done_offset = end_offset;
  if ((done_offset % BytesPerLong) != 0) {
    done_offset -= BytesPerInt;
  }
  if (done_offset > start_offset) {
    mem = clear_memory(ctl, mem, dest,
                       start_offset, phase->MakeConX(done_offset), phase);
  }
  if (done_offset < end_offset) { // emit the final 32-bit store
    Node* adr = new (C, 4) AddPNode(dest, dest, phase->MakeConX(done_offset));
    adr = phase->transform(adr);
    const TypePtr* atp = TypeRawPtr::BOTTOM;
    mem = StoreNode::make(*phase, ctl, mem, adr, atp, phase->zerocon(T_INT), T_INT);
    mem = phase->transform(mem);
    done_offset += BytesPerInt;
  }
  assert(done_offset == end_offset, "");
  return mem;
}

//=============================================================================
// Do we match on this edge? No memory edges
uint StrCompNode::match_edge(uint idx) const {
  return idx == 5 || idx == 6;
}

//------------------------------Ideal------------------------------------------
// Return a node which is more "ideal" than the current node.  Strip out
// control copies
Node *StrCompNode::Ideal(PhaseGVN *phase, bool can_reshape){
  return remove_dead_region(phase, can_reshape) ? this : NULL;
}


//=============================================================================
MemBarNode::MemBarNode(Compile* C, int alias_idx, Node* precedent)
  : MultiNode(TypeFunc::Parms + (precedent == NULL? 0: 1)),
    _adr_type(C->get_adr_type(alias_idx))
{
  init_class_id(Class_MemBar);
  Node* top = C->top();
  init_req(TypeFunc::I_O,top);
  init_req(TypeFunc::FramePtr,top);
  init_req(TypeFunc::ReturnAdr,top);
  if (precedent != NULL)
    init_req(TypeFunc::Parms, precedent);
}

//------------------------------cmp--------------------------------------------
uint MemBarNode::hash() const { return NO_HASH; }
uint MemBarNode::cmp( const Node &n ) const {
  return (&n == this);          // Always fail except on self
}

//------------------------------make-------------------------------------------
MemBarNode* MemBarNode::make(Compile* C, int opcode, int atp, Node* pn) {
  int len = Precedent + (pn == NULL? 0: 1);
  switch (opcode) {
  case Op_MemBarAcquire:   return new(C, len) MemBarAcquireNode(C,  atp, pn);
  case Op_MemBarRelease:   return new(C, len) MemBarReleaseNode(C,  atp, pn);
  case Op_MemBarVolatile:  return new(C, len) MemBarVolatileNode(C, atp, pn);
  case Op_MemBarCPUOrder:  return new(C, len) MemBarCPUOrderNode(C, atp, pn);
  case Op_Initialize:      return new(C, len) InitializeNode(C,     atp, pn);
  default:                 ShouldNotReachHere(); return NULL;
  }
}

//------------------------------Ideal------------------------------------------
// Return a node which is more "ideal" than the current node.  Strip out
// control copies
Node *MemBarNode::Ideal(PhaseGVN *phase, bool can_reshape) {
  if (remove_dead_region(phase, can_reshape))  return this;
  return NULL;
}

//------------------------------Value------------------------------------------
const Type *MemBarNode::Value( PhaseTransform *phase ) const {
  if( !in(0) ) return Type::TOP;
  if( phase->type(in(0)) == Type::TOP )
    return Type::TOP;
  return TypeTuple::MEMBAR;
}

//------------------------------match------------------------------------------
// Construct projections for memory.
Node *MemBarNode::match( const ProjNode *proj, const Matcher *m ) {
  switch (proj->_con) {
  case TypeFunc::Control:
  case TypeFunc::Memory:
    return new (m->C, 1) MachProjNode(this,proj->_con,RegMask::Empty,MachProjNode::unmatched_proj);
  }
  ShouldNotReachHere();
  return NULL;
}

//===========================InitializeNode====================================
// SUMMARY:
// This node acts as a memory barrier on raw memory, after some raw stores.
// The 'cooked' oop value feeds from the Initialize, not the Allocation.
// The Initialize can 'capture' suitably constrained stores as raw inits.
// It can coalesce related raw stores into larger units (called 'tiles').
// It can avoid zeroing new storage for memory units which have raw inits.
// At macro-expansion, it is marked 'complete', and does not optimize further.
//
// EXAMPLE:
// The object 'new short[2]' occupies 16 bytes in a 32-bit machine.
//   ctl = incoming control; mem* = incoming memory
// (Note:  A star * on a memory edge denotes I/O and other standard edges.)
// First allocate uninitialized memory and fill in the header:
//   alloc = (Allocate ctl mem* 16 #short[].klass ...)
//   ctl := alloc.Control; mem* := alloc.Memory*
//   rawmem = alloc.Memory; rawoop = alloc.RawAddress
// Then initialize to zero the non-header parts of the raw memory block:
//   init = (Initialize alloc.Control alloc.Memory* alloc.RawAddress)
//   ctl := init.Control; mem.SLICE(#short[*]) := init.Memory
// After the initialize node executes, the object is ready for service:
//   oop := (CheckCastPP init.Control alloc.RawAddress #short[])
// Suppose its body is immediately initialized as {1,2}:
//   store1 = (StoreC init.Control init.Memory (+ oop 12) 1)
//   store2 = (StoreC init.Control store1      (+ oop 14) 2)
//   mem.SLICE(#short[*]) := store2
//
// DETAILS:
// An InitializeNode collects and isolates object initialization after
// an AllocateNode and before the next possible safepoint.  As a
// memory barrier (MemBarNode), it keeps critical stores from drifting
// down past any safepoint or any publication of the allocation.
// Before this barrier, a newly-allocated object may have uninitialized bits.
// After this barrier, it may be treated as a real oop, and GC is allowed.
//
// The semantics of the InitializeNode include an implicit zeroing of
// the new object from object header to the end of the object.
// (The object header and end are determined by the AllocateNode.)
//
// Certain stores may be added as direct inputs to the InitializeNode.
// These stores must update raw memory, and they must be to addresses
// derived from the raw address produced by AllocateNode, and with
// a constant offset.  They must be ordered by increasing offset.
// The first one is at in(RawStores), the last at in(req()-1).
// Unlike most memory operations, they are not linked in a chain,
// but are displayed in parallel as users of the rawmem output of
// the allocation.
//
// (See comments in InitializeNode::capture_store, which continue
// the example given above.)
//
// When the associated Allocate is macro-expanded, the InitializeNode
// may be rewritten to optimize collected stores.  A ClearArrayNode
// may also be created at that point to represent any required zeroing.
// The InitializeNode is then marked 'complete', prohibiting further
// capturing of nearby memory operations.
//
// During macro-expansion, all captured initializations which store
// constant values of 32 bits or smaller are coalesced (if advantagous)
// into larger 'tiles' 32 or 64 bits.  This allows an object to be
// initialized in fewer memory operations.  Memory words which are
// covered by neither tiles nor non-constant stores are pre-zeroed
// by explicit stores of zero.  (The code shape happens to do all
// zeroing first, then all other stores, with both sequences occurring
// in order of ascending offsets.)
//
// Alternatively, code may be inserted between an AllocateNode and its
// InitializeNode, to perform arbitrary initialization of the new object.
// E.g., the object copying intrinsics insert complex data transfers here.
// The initialization must then be marked as 'complete' disable the
// built-in zeroing semantics and the collection of initializing stores.
//
// While an InitializeNode is incomplete, reads from the memory state
// produced by it are optimizable if they match the control edge and
// new oop address associated with the allocation/initialization.
// They return a stored value (if the offset matches) or else zero.
// A write to the memory state, if it matches control and address,
// and if it is to a constant offset, may be 'captured' by the
// InitializeNode.  It is cloned as a raw memory operation and rewired
// inside the initialization, to the raw oop produced by the allocation.
// Operations on addresses which are provably distinct (e.g., to
// other AllocateNodes) are allowed to bypass the initialization.
//
// The effect of all this is to consolidate object initialization
// (both arrays and non-arrays, both piecewise and bulk) into a
// single location, where it can be optimized as a unit.
//
// Only stores with an offset less than TrackedInitializationLimit words
// will be considered for capture by an InitializeNode.  This puts a
// reasonable limit on the complexity of optimized initializations.

//---------------------------InitializeNode------------------------------------
InitializeNode::InitializeNode(Compile* C, int adr_type, Node* rawoop)
  : _is_complete(false),
    MemBarNode(C, adr_type, rawoop)
{
  init_class_id(Class_Initialize);

  assert(adr_type == Compile::AliasIdxRaw, "only valid atp");
  assert(in(RawAddress) == rawoop, "proper init");
  // Note:  allocation() can be NULL, for secondary initialization barriers
}

// Since this node is not matched, it will be processed by the
// register allocator.  Declare that there are no constraints
// on the allocation of the RawAddress edge.
const RegMask &InitializeNode::in_RegMask(uint idx) const {
  // This edge should be set to top, by the set_complete.  But be conservative.
  if (idx == InitializeNode::RawAddress)
    return *(Compile::current()->matcher()->idealreg2spillmask[in(idx)->ideal_reg()]);
  return RegMask::Empty;
}

Node* InitializeNode::memory(uint alias_idx) {
  Node* mem = in(Memory);
  if (mem->is_MergeMem()) {
    return mem->as_MergeMem()->memory_at(alias_idx);
  } else {
    // incoming raw memory is not split
    return mem;
  }
}

bool InitializeNode::is_non_zero() {
  if (is_complete())  return false;
  remove_extra_zeroes();
  return (req() > RawStores);
}

void InitializeNode::set_complete(PhaseGVN* phase) {
  assert(!is_complete(), "caller responsibility");
  _is_complete = true;

  // After this node is complete, it contains a bunch of
  // raw-memory initializations.  There is no need for
  // it to have anything to do with non-raw memory effects.
  // Therefore, tell all non-raw users to re-optimize themselves,
  // after skipping the memory effects of this initialization.
  PhaseIterGVN* igvn = phase->is_IterGVN();
  if (igvn)  igvn->add_users_to_worklist(this);
}

// convenience function
// return false if the init contains any stores already
bool AllocateNode::maybe_set_complete(PhaseGVN* phase) {
  InitializeNode* init = initialization();
  if (init == NULL || init->is_complete())  return false;
  init->remove_extra_zeroes();
  // for now, if this allocation has already collected any inits, bail:
  if (init->is_non_zero())  return false;
  init->set_complete(phase);
  return true;
}

void InitializeNode::remove_extra_zeroes() {
  if (req() == RawStores)  return;
  Node* zmem = zero_memory();
  uint fill = RawStores;
  for (uint i = fill; i < req(); i++) {
    Node* n = in(i);
    if (n->is_top() || n == zmem)  continue;  // skip
    if (fill < i)  set_req(fill, n);          // compact
    ++fill;
  }
  // delete any empty spaces created:
  while (fill < req()) {
    del_req(fill);
  }
}

// Helper for remembering which stores go with which offsets.
intptr_t InitializeNode::get_store_offset(Node* st, PhaseTransform* phase) {
  if (!st->is_Store())  return -1;  // can happen to dead code via subsume_node
  intptr_t offset = -1;
  Node* base = AddPNode::Ideal_base_and_offset(st->in(MemNode::Address),
                                               phase, offset);
  if (base == NULL)     return -1;  // something is dead,
  if (offset < 0)       return -1;  //        dead, dead
  return offset;
}

// Helper for proving that an initialization expression is
// "simple enough" to be folded into an object initialization.
// Attempts to prove that a store's initial value 'n' can be captured
// within the initialization without creating a vicious cycle, such as:
//     { Foo p = new Foo(); p.next = p; }
// True for constants and parameters and small combinations thereof.
bool InitializeNode::detect_init_independence(Node* n,
                                              bool st_is_pinned,
                                              int& count) {
  if (n == NULL)      return true;   // (can this really happen?)
  if (n->is_Proj())   n = n->in(0);
  if (n == this)      return false;  // found a cycle
  if (n->is_Con())    return true;
  if (n->is_Start())  return true;   // params, etc., are OK
  if (n->is_Root())   return true;   // even better

  Node* ctl = n->in(0);
  if (ctl != NULL && !ctl->is_top()) {
    if (ctl->is_Proj())  ctl = ctl->in(0);
    if (ctl == this)  return false;

    // If we already know that the enclosing memory op is pinned right after
    // the init, then any control flow that the store has picked up
    // must have preceded the init, or else be equal to the init.
    // Even after loop optimizations (which might change control edges)
    // a store is never pinned *before* the availability of its inputs.
    if (!MemNode::detect_dominating_control(ctl, this->in(0)))
      return false;                  // failed to prove a good control

  }

  // Check data edges for possible dependencies on 'this'.
  if ((count += 1) > 20)  return false;  // complexity limit
  for (uint i = 1; i < n->req(); i++) {
    Node* m = n->in(i);
    if (m == NULL || m == n || m->is_top())  continue;
    uint first_i = n->find_edge(m);
    if (i != first_i)  continue;  // process duplicate edge just once
    if (!detect_init_independence(m, st_is_pinned, count)) {
      return false;
    }
  }

  return true;
}

// Here are all the checks a Store must pass before it can be moved into
// an initialization.  Returns zero if a check fails.
// On success, returns the (constant) offset to which the store applies,
// within the initialized memory.
intptr_t InitializeNode::can_capture_store(StoreNode* st, PhaseTransform* phase) {
  const int FAIL = 0;
  if (st->req() != MemNode::ValueIn + 1)
    return FAIL;                // an inscrutable StoreNode (card mark?)
  Node* ctl = st->in(MemNode::Control);
  if (!(ctl != NULL && ctl->is_Proj() && ctl->in(0) == this))
    return FAIL;                // must be unconditional after the initialization
  Node* mem = st->in(MemNode::Memory);
  if (!(mem->is_Proj() && mem->in(0) == this))
    return FAIL;                // must not be preceded by other stores
  Node* adr = st->in(MemNode::Address);
  intptr_t offset;
  AllocateNode* alloc = AllocateNode::Ideal_allocation(adr, phase, offset);
  if (alloc == NULL)
    return FAIL;                // inscrutable address
  if (alloc != allocation())
    return FAIL;                // wrong allocation!  (store needs to float up)
  Node* val = st->in(MemNode::ValueIn);
  int complexity_count = 0;
  if (!detect_init_independence(val, true, complexity_count))
    return FAIL;                // stored value must be 'simple enough'

  return offset;                // success
}

// Find the captured store in(i) which corresponds to the range
// [start..start+size) in the initialized object.
// If there is one, return its index i.  If there isn't, return the
// negative of the index where it should be inserted.
// Return 0 if the queried range overlaps an initialization boundary
// or if dead code is encountered.
// If size_in_bytes is zero, do not bother with overlap checks.
int InitializeNode::captured_store_insertion_point(intptr_t start,
                                                   int size_in_bytes,
                                                   PhaseTransform* phase) {
  const int FAIL = 0, MAX_STORE = BytesPerLong;

  if (is_complete())
    return FAIL;                // arraycopy got here first; punt

  assert(allocation() != NULL, "must be present");

  // no negatives, no header fields:
  if (start < (intptr_t) allocation()->minimum_header_size())  return FAIL;

  // after a certain size, we bail out on tracking all the stores:
  intptr_t ti_limit = (TrackedInitializationLimit * HeapWordSize);
  if (start >= ti_limit)  return FAIL;

  for (uint i = InitializeNode::RawStores, limit = req(); ; ) {
    if (i >= limit)  return -(int)i; // not found; here is where to put it

    Node*    st     = in(i);
    intptr_t st_off = get_store_offset(st, phase);
    if (st_off < 0) {
      if (st != zero_memory()) {
        return FAIL;            // bail out if there is dead garbage
      }
    } else if (st_off > start) {
      // ...we are done, since stores are ordered
      if (st_off < start + size_in_bytes) {
        return FAIL;            // the next store overlaps
      }
      return -(int)i;           // not found; here is where to put it
    } else if (st_off < start) {
      if (size_in_bytes != 0 &&
          start < st_off + MAX_STORE &&
          start < st_off + st->as_Store()->memory_size()) {
        return FAIL;            // the previous store overlaps
      }
    } else {
      if (size_in_bytes != 0 &&
          st->as_Store()->memory_size() != size_in_bytes) {
        return FAIL;            // mismatched store size
      }
      return i;
    }

    ++i;
  }
}

// Look for a captured store which initializes at the offset 'start'
// with the given size.  If there is no such store, and no other
// initialization interferes, then return zero_memory (the memory
// projection of the AllocateNode).
Node* InitializeNode::find_captured_store(intptr_t start, int size_in_bytes,
                                          PhaseTransform* phase) {
  assert(stores_are_sane(phase), "");
  int i = captured_store_insertion_point(start, size_in_bytes, phase);
  if (i == 0) {
    return NULL;                // something is dead
  } else if (i < 0) {
    return zero_memory();       // just primordial zero bits here
  } else {
    Node* st = in(i);           // here is the store at this position
    assert(get_store_offset(st->as_Store(), phase) == start, "sanity");
    return st;
  }
}

// Create, as a raw pointer, an address within my new object at 'offset'.
Node* InitializeNode::make_raw_address(intptr_t offset,
                                       PhaseTransform* phase) {
  Node* addr = in(RawAddress);
  if (offset != 0) {
    Compile* C = phase->C;
    addr = phase->transform( new (C, 4) AddPNode(C->top(), addr,
                                                 phase->MakeConX(offset)) );
  }
  return addr;
}

// Clone the given store, converting it into a raw store
// initializing a field or element of my new object.
// Caller is responsible for retiring the original store,
// with subsume_node or the like.
//
// From the example above InitializeNode::InitializeNode,
// here are the old stores to be captured:
//   store1 = (StoreC init.Control init.Memory (+ oop 12) 1)
//   store2 = (StoreC init.Control store1      (+ oop 14) 2)
//
// Here is the changed code; note the extra edges on init:
//   alloc = (Allocate ...)
//   rawoop = alloc.RawAddress
//   rawstore1 = (StoreC alloc.Control alloc.Memory (+ rawoop 12) 1)
//   rawstore2 = (StoreC alloc.Control alloc.Memory (+ rawoop 14) 2)
//   init = (Initialize alloc.Control alloc.Memory rawoop
//                      rawstore1 rawstore2)
//
Node* InitializeNode::capture_store(StoreNode* st, intptr_t start,
                                    PhaseTransform* phase) {
  assert(stores_are_sane(phase), "");

  if (start < 0)  return NULL;
  assert(can_capture_store(st, phase) == start, "sanity");

  Compile* C = phase->C;
  int size_in_bytes = st->memory_size();
  int i = captured_store_insertion_point(start, size_in_bytes, phase);
  if (i == 0)  return NULL;     // bail out
  Node* prev_mem = NULL;        // raw memory for the captured store
  if (i > 0) {
    prev_mem = in(i);           // there is a pre-existing store under this one
    set_req(i, C->top());       // temporarily disconnect it
    // See StoreNode::Ideal 'st->outcnt() == 1' for the reason to disconnect.
  } else {
    i = -i;                     // no pre-existing store
    prev_mem = zero_memory();   // a slice of the newly allocated object
    if (i > InitializeNode::RawStores && in(i-1) == prev_mem)
      set_req(--i, C->top());   // reuse this edge; it has been folded away
    else
      ins_req(i, C->top());     // build a new edge
  }
  Node* new_st = st->clone();
  new_st->set_req(MemNode::Control, in(Control));
  new_st->set_req(MemNode::Memory,  prev_mem);
  new_st->set_req(MemNode::Address, make_raw_address(start, phase));
  new_st = phase->transform(new_st);

  // At this point, new_st might have swallowed a pre-existing store
  // at the same offset, or perhaps new_st might have disappeared,
  // if it redundantly stored the same value (or zero to fresh memory).

  // In any case, wire it in:
  set_req(i, new_st);

  // The caller may now kill the old guy.
  DEBUG_ONLY(Node* check_st = find_captured_store(start, size_in_bytes, phase));
  assert(check_st == new_st || check_st == NULL, "must be findable");
  assert(!is_complete(), "");
  return new_st;
}

static bool store_constant(jlong* tiles, int num_tiles,
                           intptr_t st_off, int st_size,
                           jlong con) {
  if ((st_off & (st_size-1)) != 0)
    return false;               // strange store offset (assume size==2**N)
  address addr = (address)tiles + st_off;
  assert(st_off >= 0 && addr+st_size <= (address)&tiles[num_tiles], "oob");
  switch (st_size) {
  case sizeof(jbyte):  *(jbyte*) addr = (jbyte) con; break;
  case sizeof(jchar):  *(jchar*) addr = (jchar) con; break;
  case sizeof(jint):   *(jint*)  addr = (jint)  con; break;
  case sizeof(jlong):  *(jlong*) addr = (jlong) con; break;
  default: return false;        // strange store size (detect size!=2**N here)
  }
  return true;                  // return success to caller
}

// Coalesce subword constants into int constants and possibly
// into long constants.  The goal, if the CPU permits,
// is to initialize the object with a small number of 64-bit tiles.
// Also, convert floating-point constants to bit patterns.
// Non-constants are not relevant to this pass.
//
// In terms of the running example on InitializeNode::InitializeNode
// and InitializeNode::capture_store, here is the transformation
// of rawstore1 and rawstore2 into rawstore12:
//   alloc = (Allocate ...)
//   rawoop = alloc.RawAddress
//   tile12 = 0x00010002
//   rawstore12 = (StoreI alloc.Control alloc.Memory (+ rawoop 12) tile12)
//   init = (Initialize alloc.Control alloc.Memory rawoop rawstore12)
//
void
InitializeNode::coalesce_subword_stores(intptr_t header_size,
                                        Node* size_in_bytes,
                                        PhaseGVN* phase) {
  Compile* C = phase->C;

  assert(stores_are_sane(phase), "");
  // Note:  After this pass, they are not completely sane,
  // since there may be some overlaps.

  int old_subword = 0, old_long = 0, new_int = 0, new_long = 0;

  intptr_t ti_limit = (TrackedInitializationLimit * HeapWordSize);
  intptr_t size_limit = phase->find_intptr_t_con(size_in_bytes, ti_limit);
  size_limit = MIN2(size_limit, ti_limit);
  size_limit = align_size_up(size_limit, BytesPerLong);
  int num_tiles = size_limit / BytesPerLong;

  // allocate space for the tile map:
  const int small_len = DEBUG_ONLY(true ? 3 :) 30; // keep stack frames small
  jlong  tiles_buf[small_len];
  Node*  nodes_buf[small_len];
  jlong  inits_buf[small_len];
  jlong* tiles = ((num_tiles <= small_len) ? &tiles_buf[0]
                  : NEW_RESOURCE_ARRAY(jlong, num_tiles));
  Node** nodes = ((num_tiles <= small_len) ? &nodes_buf[0]
                  : NEW_RESOURCE_ARRAY(Node*, num_tiles));
  jlong* inits = ((num_tiles <= small_len) ? &inits_buf[0]
                  : NEW_RESOURCE_ARRAY(jlong, num_tiles));
  // tiles: exact bitwise model of all primitive constants
  // nodes: last constant-storing node subsumed into the tiles model
  // inits: which bytes (in each tile) are touched by any initializations

  //// Pass A: Fill in the tile model with any relevant stores.

  Copy::zero_to_bytes(tiles, sizeof(tiles[0]) * num_tiles);
  Copy::zero_to_bytes(nodes, sizeof(nodes[0]) * num_tiles);
  Copy::zero_to_bytes(inits, sizeof(inits[0]) * num_tiles);
  Node* zmem = zero_memory(); // initially zero memory state
  for (uint i = InitializeNode::RawStores, limit = req(); i < limit; i++) {
    Node* st = in(i);
    intptr_t st_off = get_store_offset(st, phase);

    // Figure out the store's offset and constant value:
    if (st_off < header_size)             continue; //skip (ignore header)
    if (st->in(MemNode::Memory) != zmem)  continue; //skip (odd store chain)
    int st_size = st->as_Store()->memory_size();
    if (st_off + st_size > size_limit)    break;

    // Record which bytes are touched, whether by constant or not.
    if (!store_constant(inits, num_tiles, st_off, st_size, (jlong) -1))
      continue;                 // skip (strange store size)

    const Type* val = phase->type(st->in(MemNode::ValueIn));
    if (!val->singleton())                continue; //skip (non-con store)
    BasicType type = val->basic_type();

    jlong con = 0;
    switch (type) {
    case T_INT:    con = val->is_int()->get_con();  break;
    case T_LONG:   con = val->is_long()->get_con(); break;
    case T_FLOAT:  con = jint_cast(val->getf());    break;
    case T_DOUBLE: con = jlong_cast(val->getd());   break;
    default:                              continue; //skip (odd store type)
    }

    if (type == T_LONG && Matcher::isSimpleConstant64(con) &&
        st->Opcode() == Op_StoreL) {
      continue;                 // This StoreL is already optimal.
    }

    // Store down the constant.
    store_constant(tiles, num_tiles, st_off, st_size, con);

    intptr_t j = st_off >> LogBytesPerLong;

    if (type == T_INT && st_size == BytesPerInt
        && (st_off & BytesPerInt) == BytesPerInt) {
      jlong lcon = tiles[j];
      if (!Matcher::isSimpleConstant64(lcon) &&
          st->Opcode() == Op_StoreI) {
        // This StoreI is already optimal by itself.
        jint* intcon = (jint*) &tiles[j];
        intcon[1] = 0;  // undo the store_constant()

        // If the previous store is also optimal by itself, back up and
        // undo the action of the previous loop iteration... if we can.
        // But if we can't, just let the previous half take care of itself.
        st = nodes[j];
        st_off -= BytesPerInt;
        con = intcon[0];
        if (con != 0 && st != NULL && st->Opcode() == Op_StoreI) {
          assert(st_off >= header_size, "still ignoring header");
          assert(get_store_offset(st, phase) == st_off, "must be");
          assert(in(i-1) == zmem, "must be");
          DEBUG_ONLY(const Type* tcon = phase->type(st->in(MemNode::ValueIn)));
          assert(con == tcon->is_int()->get_con(), "must be");
          // Undo the effects of the previous loop trip, which swallowed st:
          intcon[0] = 0;        // undo store_constant()
          set_req(i-1, st);     // undo set_req(i, zmem)
          nodes[j] = NULL;      // undo nodes[j] = st
          --old_subword;        // undo ++old_subword
        }
        continue;               // This StoreI is already optimal.
      }
    }

    // This store is not needed.
    set_req(i, zmem);
    nodes[j] = st;              // record for the moment
    if (st_size < BytesPerLong) // something has changed
          ++old_subword;        // includes int/float, but who's counting...
    else  ++old_long;
  }

  if ((old_subword + old_long) == 0)
    return;                     // nothing more to do

  //// Pass B: Convert any non-zero tiles into optimal constant stores.
  // Be sure to insert them before overlapping non-constant stores.
  // (E.g., byte[] x = { 1,2,y,4 }  =>  x[int 0] = 0x01020004, x[2]=y.)
  for (int j = 0; j < num_tiles; j++) {
    jlong con  = tiles[j];
    jlong init = inits[j];
    if (con == 0)  continue;
    jint con0,  con1;           // split the constant, address-wise
    jint init0, init1;          // split the init map, address-wise
    { union { jlong con; jint intcon[2]; } u;
      u.con = con;
      con0  = u.intcon[0];
      con1  = u.intcon[1];
      u.con = init;
      init0 = u.intcon[0];
      init1 = u.intcon[1];
    }

    Node* old = nodes[j];
    assert(old != NULL, "need the prior store");
    intptr_t offset = (j * BytesPerLong);

    bool split = !Matcher::isSimpleConstant64(con);

    if (offset < header_size) {
      assert(offset + BytesPerInt >= header_size, "second int counts");
      assert(*(jint*)&tiles[j] == 0, "junk in header");
      split = true;             // only the second word counts
      // Example:  int a[] = { 42 ... }
    } else if (con0 == 0 && init0 == -1) {
      split = true;             // first word is covered by full inits
      // Example:  int a[] = { ... foo(), 42 ... }
    } else if (con1 == 0 && init1 == -1) {
      split = true;             // second word is covered by full inits
      // Example:  int a[] = { ... 42, foo() ... }
    }

    // Here's a case where init0 is neither 0 nor -1:
    //   byte a[] = { ... 0,0,foo(),0,  0,0,0,42 ... }
    // Assuming big-endian memory, init0, init1 are 0x0000FF00, 0x000000FF.
    // In this case the tile is not split; it is (jlong)42.
    // The big tile is stored down, and then the foo() value is inserted.
    // (If there were foo(),foo() instead of foo(),0, init0 would be -1.)

    Node* ctl = old->in(MemNode::Control);
    Node* adr = make_raw_address(offset, phase);
    const TypePtr* atp = TypeRawPtr::BOTTOM;

    // One or two coalesced stores to plop down.
    Node*    st[2];
    intptr_t off[2];
    int  nst = 0;
    if (!split) {
      ++new_long;
      off[nst] = offset;
      st[nst++] = StoreNode::make(*phase, ctl, zmem, adr, atp,
                                  phase->longcon(con), T_LONG);
    } else {
      // Omit either if it is a zero.
      if (con0 != 0) {
        ++new_int;
        off[nst]  = offset;
        st[nst++] = StoreNode::make(*phase, ctl, zmem, adr, atp,
                                    phase->intcon(con0), T_INT);
      }
      if (con1 != 0) {
        ++new_int;
        offset += BytesPerInt;
        adr = make_raw_address(offset, phase);
        off[nst]  = offset;
        st[nst++] = StoreNode::make(*phase, ctl, zmem, adr, atp,
                                    phase->intcon(con1), T_INT);
      }
    }

    // Insert second store first, then the first before the second.
    // Insert each one just before any overlapping non-constant stores.
    while (nst > 0) {
      Node* st1 = st[--nst];
      C->copy_node_notes_to(st1, old);
      st1 = phase->transform(st1);
      offset = off[nst];
      assert(offset >= header_size, "do not smash header");
      int ins_idx = captured_store_insertion_point(offset, /*size:*/0, phase);
      guarantee(ins_idx != 0, "must re-insert constant store");
      if (ins_idx < 0)  ins_idx = -ins_idx;  // never overlap
      if (ins_idx > InitializeNode::RawStores && in(ins_idx-1) == zmem)
        set_req(--ins_idx, st1);
      else
        ins_req(ins_idx, st1);
    }
  }

  if (PrintCompilation && WizardMode)
    tty->print_cr("Changed %d/%d subword/long constants into %d/%d int/long",
                  old_subword, old_long, new_int, new_long);
  if (C->log() != NULL)
    C->log()->elem("comment that='%d/%d subword/long to %d/%d int/long'",
                   old_subword, old_long, new_int, new_long);

  // Clean up any remaining occurrences of zmem:
  remove_extra_zeroes();
}

// Explore forward from in(start) to find the first fully initialized
// word, and return its offset.  Skip groups of subword stores which
// together initialize full words.  If in(start) is itself part of a
// fully initialized word, return the offset of in(start).  If there
// are no following full-word stores, or if something is fishy, return
// a negative value.
intptr_t InitializeNode::find_next_fullword_store(uint start, PhaseGVN* phase) {
  int       int_map = 0;
  intptr_t  int_map_off = 0;
  const int FULL_MAP = right_n_bits(BytesPerInt);  // the int_map we hope for

  for (uint i = start, limit = req(); i < limit; i++) {
    Node* st = in(i);

    intptr_t st_off = get_store_offset(st, phase);
    if (st_off < 0)  break;  // return conservative answer

    int st_size = st->as_Store()->memory_size();
    if (st_size >= BytesPerInt && (st_off % BytesPerInt) == 0) {
      return st_off;            // we found a complete word init
    }

    // update the map:

    intptr_t this_int_off = align_size_down(st_off, BytesPerInt);
    if (this_int_off != int_map_off) {
      // reset the map:
      int_map = 0;
      int_map_off = this_int_off;
    }

    int subword_off = st_off - this_int_off;
    int_map |= right_n_bits(st_size) << subword_off;
    if ((int_map & FULL_MAP) == FULL_MAP) {
      return this_int_off;      // we found a complete word init
    }

    // Did this store hit or cross the word boundary?
    intptr_t next_int_off = align_size_down(st_off + st_size, BytesPerInt);
    if (next_int_off == this_int_off + BytesPerInt) {
      // We passed the current int, without fully initializing it.
      int_map_off = next_int_off;
      int_map >>= BytesPerInt;
    } else if (next_int_off > this_int_off + BytesPerInt) {
      // We passed the current and next int.
      return this_int_off + BytesPerInt;
    }
  }

  return -1;
}


// Called when the associated AllocateNode is expanded into CFG.
// At this point, we may perform additional optimizations.
// Linearize the stores by ascending offset, to make memory
// activity as coherent as possible.
Node* InitializeNode::complete_stores(Node* rawctl, Node* rawmem, Node* rawptr,
                                      intptr_t header_size,
                                      Node* size_in_bytes,
                                      PhaseGVN* phase) {
  assert(!is_complete(), "not already complete");
  assert(stores_are_sane(phase), "");
  assert(allocation() != NULL, "must be present");

  remove_extra_zeroes();

  if (ReduceFieldZeroing || ReduceBulkZeroing)
    // reduce instruction count for common initialization patterns
    coalesce_subword_stores(header_size, size_in_bytes, phase);

  Node* zmem = zero_memory();   // initially zero memory state
  Node* inits = zmem;           // accumulating a linearized chain of inits
  #ifdef ASSERT
  intptr_t first_offset = allocation()->minimum_header_size();
  intptr_t last_init_off = first_offset;  // previous init offset
  intptr_t last_init_end = first_offset;  // previous init offset+size
  intptr_t last_tile_end = first_offset;  // previous tile offset+size
  #endif
  intptr_t zeroes_done = header_size;

  bool do_zeroing = true;       // we might give up if inits are very sparse
  int  big_init_gaps = 0;       // how many large gaps have we seen?

  if (ZeroTLAB)  do_zeroing = false;
  if (!ReduceFieldZeroing && !ReduceBulkZeroing)  do_zeroing = false;

  for (uint i = InitializeNode::RawStores, limit = req(); i < limit; i++) {
    Node* st = in(i);
    intptr_t st_off = get_store_offset(st, phase);
    if (st_off < 0)
      break;                    // unknown junk in the inits
    if (st->in(MemNode::Memory) != zmem)
      break;                    // complicated store chains somehow in list

    int st_size = st->as_Store()->memory_size();
    intptr_t next_init_off = st_off + st_size;

    if (do_zeroing && zeroes_done < next_init_off) {
      // See if this store needs a zero before it or under it.
      intptr_t zeroes_needed = st_off;

      if (st_size < BytesPerInt) {
        // Look for subword stores which only partially initialize words.
        // If we find some, we must lay down some word-level zeroes first,
        // underneath the subword stores.
        //
        // Examples:
        //   byte[] a = { p,q,r,s }  =>  a[0]=p,a[1]=q,a[2]=r,a[3]=s
        //   byte[] a = { x,y,0,0 }  =>  a[0..3] = 0, a[0]=x,a[1]=y
        //   byte[] a = { 0,0,z,0 }  =>  a[0..3] = 0, a[2]=z
        //
        // Note:  coalesce_subword_stores may have already done this,
        // if it was prompted by constant non-zero subword initializers.
        // But this case can still arise with non-constant stores.

        intptr_t next_full_store = find_next_fullword_store(i, phase);

        // In the examples above:
        //   in(i)          p   q   r   s     x   y     z
        //   st_off        12  13  14  15    12  13    14
        //   st_size        1   1   1   1     1   1     1
        //   next_full_s.  12  16  16  16    16  16    16
        //   z's_done      12  16  16  16    12  16    12
        //   z's_needed    12  16  16  16    16  16    16
        //   zsize          0   0   0   0     4   0     4
        if (next_full_store < 0) {
          // Conservative tack:  Zero to end of current word.
          zeroes_needed = align_size_up(zeroes_needed, BytesPerInt);
        } else {
          // Zero to beginning of next fully initialized word.
          // Or, don't zero at all, if we are already in that word.
          assert(next_full_store >= zeroes_needed, "must go forward");
          assert((next_full_store & (BytesPerInt-1)) == 0, "even boundary");
          zeroes_needed = next_full_store;
        }
      }

      if (zeroes_needed > zeroes_done) {
        intptr_t zsize = zeroes_needed - zeroes_done;
        // Do some incremental zeroing on rawmem, in parallel with inits.
        zeroes_done = align_size_down(zeroes_done, BytesPerInt);
        rawmem = ClearArrayNode::clear_memory(rawctl, rawmem, rawptr,
                                              zeroes_done, zeroes_needed,
                                              phase);
        zeroes_done = zeroes_needed;
        if (zsize > Matcher::init_array_short_size && ++big_init_gaps > 2)
          do_zeroing = false;   // leave the hole, next time
      }
    }

    // Collect the store and move on:
    st->set_req(MemNode::Memory, inits);
    inits = st;                 // put it on the linearized chain
    set_req(i, zmem);           // unhook from previous position

    if (zeroes_done == st_off)
      zeroes_done = next_init_off;

    assert(!do_zeroing || zeroes_done >= next_init_off, "don't miss any");

    #ifdef ASSERT
    // Various order invariants.  Weaker than stores_are_sane because
    // a large constant tile can be filled in by smaller non-constant stores.
    assert(st_off >= last_init_off, "inits do not reverse");
    last_init_off = st_off;
    const Type* val = NULL;
    if (st_size >= BytesPerInt &&
        (val = phase->type(st->in(MemNode::ValueIn)))->singleton() &&
        (int)val->basic_type() < (int)T_OBJECT) {
      assert(st_off >= last_tile_end, "tiles do not overlap");
      assert(st_off >= last_init_end, "tiles do not overwrite inits");
      last_tile_end = MAX2(last_tile_end, next_init_off);
    } else {
      intptr_t st_tile_end = align_size_up(next_init_off, BytesPerLong);
      assert(st_tile_end >= last_tile_end, "inits stay with tiles");
      assert(st_off      >= last_init_end, "inits do not overlap");
      last_init_end = next_init_off;  // it's a non-tile
    }
    #endif //ASSERT
  }

  remove_extra_zeroes();        // clear out all the zmems left over
  add_req(inits);

  if (!ZeroTLAB) {
    // If anything remains to be zeroed, zero it all now.
    zeroes_done = align_size_down(zeroes_done, BytesPerInt);
    // if it is the last unused 4 bytes of an instance, forget about it
    intptr_t size_limit = phase->find_intptr_t_con(size_in_bytes, max_jint);
    if (zeroes_done + BytesPerLong >= size_limit) {
      assert(allocation() != NULL, "");
      Node* klass_node = allocation()->in(AllocateNode::KlassNode);
      ciKlass* k = phase->type(klass_node)->is_klassptr()->klass();
      if (zeroes_done == k->layout_helper())
        zeroes_done = size_limit;
    }
    if (zeroes_done < size_limit) {
      rawmem = ClearArrayNode::clear_memory(rawctl, rawmem, rawptr,
                                            zeroes_done, size_in_bytes, phase);
    }
  }

  set_complete(phase);
  return rawmem;
}


#ifdef ASSERT
bool InitializeNode::stores_are_sane(PhaseTransform* phase) {
  if (is_complete())
    return true;                // stores could be anything at this point
  assert(allocation() != NULL, "must be present");
  intptr_t last_off = allocation()->minimum_header_size();
  for (uint i = InitializeNode::RawStores; i < req(); i++) {
    Node* st = in(i);
    intptr_t st_off = get_store_offset(st, phase);
    if (st_off < 0)  continue;  // ignore dead garbage
    if (last_off > st_off) {
      tty->print_cr("*** bad store offset at %d: %d > %d", i, last_off, st_off);
      this->dump(2);
      assert(false, "ascending store offsets");
      return false;
    }
    last_off = st_off + st->as_Store()->memory_size();
  }
  return true;
}
#endif //ASSERT


//============================MergeMemNode=====================================
//
// SEMANTICS OF MEMORY MERGES:  A MergeMem is a memory state assembled from several
// contributing store or call operations.  Each contributor provides the memory
// state for a particular "alias type" (see Compile::alias_type).  For example,
// if a MergeMem has an input X for alias category #6, then any memory reference
// to alias category #6 may use X as its memory state input, as an exact equivalent
// to using the MergeMem as a whole.
//   Load<6>( MergeMem(<6>: X, ...), p ) <==> Load<6>(X,p)
//
// (Here, the <N> notation gives the index of the relevant adr_type.)
//
// In one special case (and more cases in the future), alias categories overlap.
// The special alias category "Bot" (Compile::AliasIdxBot) includes all memory
// states.  Therefore, if a MergeMem has only one contributing input W for Bot,
// it is exactly equivalent to that state W:
//   MergeMem(<Bot>: W) <==> W
//
// Usually, the merge has more than one input.  In that case, where inputs
// overlap (i.e., one is Bot), the narrower alias type determines the memory
// state for that type, and the wider alias type (Bot) fills in everywhere else:
//   Load<5>( MergeMem(<Bot>: W, <6>: X), p ) <==> Load<5>(W,p)
//   Load<6>( MergeMem(<Bot>: W, <6>: X), p ) <==> Load<6>(X,p)
//
// A merge can take a "wide" memory state as one of its narrow inputs.
// This simply means that the merge observes out only the relevant parts of
// the wide input.  That is, wide memory states arriving at narrow merge inputs
// are implicitly "filtered" or "sliced" as necessary.  (This is rare.)
//
// These rules imply that MergeMem nodes may cascade (via their <Bot> links),
// and that memory slices "leak through":
//   MergeMem(<Bot>: MergeMem(<Bot>: W, <7>: Y)) <==> MergeMem(<Bot>: W, <7>: Y)
//
// But, in such a cascade, repeated memory slices can "block the leak":
//   MergeMem(<Bot>: MergeMem(<Bot>: W, <7>: Y), <7>: Y') <==> MergeMem(<Bot>: W, <7>: Y')
//
// In the last example, Y is not part of the combined memory state of the
// outermost MergeMem.  The system must, of course, prevent unschedulable
// memory states from arising, so you can be sure that the state Y is somehow
// a precursor to state Y'.
//
//
// REPRESENTATION OF MEMORY MERGES: The indexes used to address the Node::in array
// of each MergeMemNode array are exactly the numerical alias indexes, including
// but not limited to AliasIdxTop, AliasIdxBot, and AliasIdxRaw.  The functions
// Compile::alias_type (and kin) produce and manage these indexes.
//
// By convention, the value of in(AliasIdxTop) (i.e., in(1)) is always the top node.
// (Note that this provides quick access to the top node inside MergeMem methods,
// without the need to reach out via TLS to Compile::current.)
//
// As a consequence of what was just described, a MergeMem that represents a full
// memory state has an edge in(AliasIdxBot) which is a "wide" memory state,
// containing all alias categories.
//
// MergeMem nodes never (?) have control inputs, so in(0) is NULL.
//
// All other edges in(N) (including in(AliasIdxRaw), which is in(3)) are either
// a memory state for the alias type <N>, or else the top node, meaning that
// there is no particular input for that alias type.  Note that the length of
// a MergeMem is variable, and may be extended at any time to accommodate new
// memory states at larger alias indexes.  When merges grow, they are of course
// filled with "top" in the unused in() positions.
//
// This use of top is named "empty_memory()", or "empty_mem" (no-memory) as a variable.
// (Top was chosen because it works smoothly with passes like GCM.)
//
// For convenience, we hardwire the alias index for TypeRawPtr::BOTTOM.  (It is
// the type of random VM bits like TLS references.)  Since it is always the
// first non-Bot memory slice, some low-level loops use it to initialize an
// index variable:  for (i = AliasIdxRaw; i < req(); i++).
//
//
// ACCESSORS:  There is a special accessor MergeMemNode::base_memory which returns
// the distinguished "wide" state.  The accessor MergeMemNode::memory_at(N) returns
// the memory state for alias type <N>, or (if there is no particular slice at <N>,
// it returns the base memory.  To prevent bugs, memory_at does not accept <Top>
// or <Bot> indexes.  The iterator MergeMemStream provides robust iteration over
// MergeMem nodes or pairs of such nodes, ensuring that the non-top edges are visited.
//
// %%%% We may get rid of base_memory as a separate accessor at some point; it isn't
// really that different from the other memory inputs.  An abbreviation called
// "bot_memory()" for "memory_at(AliasIdxBot)" would keep code tidy.
//
//
// PARTIAL MEMORY STATES:  During optimization, MergeMem nodes may arise that represent
// partial memory states.  When a Phi splits through a MergeMem, the copy of the Phi
// that "emerges though" the base memory will be marked as excluding the alias types
// of the other (narrow-memory) copies which "emerged through" the narrow edges:
//
//   Phi<Bot>(U, MergeMem(<Bot>: W, <8>: Y))
//     ==Ideal=>  MergeMem(<Bot>: Phi<Bot-8>(U, W), Phi<8>(U, Y))
//
// This strange "subtraction" effect is necessary to ensure IGVN convergence.
// (It is currently unimplemented.)  As you can see, the resulting merge is
// actually a disjoint union of memory states, rather than an overlay.
//

//------------------------------MergeMemNode-----------------------------------
Node* MergeMemNode::make_empty_memory() {
  Node* empty_memory = (Node*) Compile::current()->top();
  assert(empty_memory->is_top(), "correct sentinel identity");
  return empty_memory;
}

MergeMemNode::MergeMemNode(Node *new_base) : Node(1+Compile::AliasIdxRaw) {
  init_class_id(Class_MergeMem);
  // all inputs are nullified in Node::Node(int)
  // set_input(0, NULL);  // no control input

  // Initialize the edges uniformly to top, for starters.
  Node* empty_mem = make_empty_memory();
  for (uint i = Compile::AliasIdxTop; i < req(); i++) {
    init_req(i,empty_mem);
  }
  assert(empty_memory() == empty_mem, "");

  if( new_base != NULL && new_base->is_MergeMem() ) {
    MergeMemNode* mdef = new_base->as_MergeMem();
    assert(mdef->empty_memory() == empty_mem, "consistent sentinels");
    for (MergeMemStream mms(this, mdef); mms.next_non_empty2(); ) {
      mms.set_memory(mms.memory2());
    }
    assert(base_memory() == mdef->base_memory(), "");
  } else {
    set_base_memory(new_base);
  }
}

// Make a new, untransformed MergeMem with the same base as 'mem'.
// If mem is itself a MergeMem, populate the result with the same edges.
MergeMemNode* MergeMemNode::make(Compile* C, Node* mem) {
  return new(C, 1+Compile::AliasIdxRaw) MergeMemNode(mem);
}

//------------------------------cmp--------------------------------------------
uint MergeMemNode::hash() const { return NO_HASH; }
uint MergeMemNode::cmp( const Node &n ) const {
  return (&n == this);          // Always fail except on self
}

//------------------------------Identity---------------------------------------
Node* MergeMemNode::Identity(PhaseTransform *phase) {
  // Identity if this merge point does not record any interesting memory
  // disambiguations.
  Node* base_mem = base_memory();
  Node* empty_mem = empty_memory();
  if (base_mem != empty_mem) {  // Memory path is not dead?
    for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
      Node* mem = in(i);
      if (mem != empty_mem && mem != base_mem) {
        return this;            // Many memory splits; no change
      }
    }
  }
  return base_mem;              // No memory splits; ID on the one true input
}

//------------------------------Ideal------------------------------------------
// This method is invoked recursively on chains of MergeMem nodes
Node *MergeMemNode::Ideal(PhaseGVN *phase, bool can_reshape) {
  // Remove chain'd MergeMems
  //
  // This is delicate, because the each "in(i)" (i >= Raw) is interpreted
  // relative to the "in(Bot)".  Since we are patching both at the same time,
  // we have to be careful to read each "in(i)" relative to the old "in(Bot)",
  // but rewrite each "in(i)" relative to the new "in(Bot)".
  Node *progress = NULL;


  Node* old_base = base_memory();
  Node* empty_mem = empty_memory();
  if (old_base == empty_mem)
    return NULL; // Dead memory path.

  MergeMemNode* old_mbase;
  if (old_base != NULL && old_base->is_MergeMem())
    old_mbase = old_base->as_MergeMem();
  else
    old_mbase = NULL;
  Node* new_base = old_base;

  // simplify stacked MergeMems in base memory
  if (old_mbase)  new_base = old_mbase->base_memory();

  // the base memory might contribute new slices beyond my req()
  if (old_mbase)  grow_to_match(old_mbase);

  // Look carefully at the base node if it is a phi.
  PhiNode* phi_base;
  if (new_base != NULL && new_base->is_Phi())
    phi_base = new_base->as_Phi();
  else
    phi_base = NULL;

  Node*    phi_reg = NULL;
  uint     phi_len = (uint)-1;
  if (phi_base != NULL && !phi_base->is_copy()) {
    // do not examine phi if degraded to a copy
    phi_reg = phi_base->region();
    phi_len = phi_base->req();
    // see if the phi is unfinished
    for (uint i = 1; i < phi_len; i++) {
      if (phi_base->in(i) == NULL) {
        // incomplete phi; do not look at it yet!
        phi_reg = NULL;
        phi_len = (uint)-1;
        break;
      }
    }
  }

  // Note:  We do not call verify_sparse on entry, because inputs
  // can normalize to the base_memory via subsume_node or similar
  // mechanisms.  This method repairs that damage.

  assert(!old_mbase || old_mbase->is_empty_memory(empty_mem), "consistent sentinels");

  // Look at each slice.
  for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
    Node* old_in = in(i);
    // calculate the old memory value
    Node* old_mem = old_in;
    if (old_mem == empty_mem)  old_mem = old_base;
    assert(old_mem == memory_at(i), "");

    // maybe update (reslice) the old memory value

    // simplify stacked MergeMems
    Node* new_mem = old_mem;
    MergeMemNode* old_mmem;
    if (old_mem != NULL && old_mem->is_MergeMem())
      old_mmem = old_mem->as_MergeMem();
    else
      old_mmem = NULL;
    if (old_mmem == this) {
      // This can happen if loops break up and safepoints disappear.
      // A merge of BotPtr (default) with a RawPtr memory derived from a
      // safepoint can be rewritten to a merge of the same BotPtr with
      // the BotPtr phi coming into the loop.  If that phi disappears
      // also, we can end up with a self-loop of the mergemem.
      // In general, if loops degenerate and memory effects disappear,
      // a mergemem can be left looking at itself.  This simply means
      // that the mergemem's default should be used, since there is
      // no longer any apparent effect on this slice.
      // Note: If a memory slice is a MergeMem cycle, it is unreachable
      //       from start.  Update the input to TOP.
      new_mem = (new_base == this || new_base == empty_mem)? empty_mem : new_base;
    }
    else if (old_mmem != NULL) {
      new_mem = old_mmem->memory_at(i);
    }
    // else preceeding memory was not a MergeMem

    // replace equivalent phis (unfortunately, they do not GVN together)
    if (new_mem != NULL && new_mem != new_base &&
        new_mem->req() == phi_len && new_mem->in(0) == phi_reg) {
      if (new_mem->is_Phi()) {
        PhiNode* phi_mem = new_mem->as_Phi();
        for (uint i = 1; i < phi_len; i++) {
          if (phi_base->in(i) != phi_mem->in(i)) {
            phi_mem = NULL;
            break;
          }
        }
        if (phi_mem != NULL) {
          // equivalent phi nodes; revert to the def
          new_mem = new_base;
        }
      }
    }

    // maybe store down a new value
    Node* new_in = new_mem;
    if (new_in == new_base)  new_in = empty_mem;

    if (new_in != old_in) {
      // Warning:  Do not combine this "if" with the previous "if"
      // A memory slice might have be be rewritten even if it is semantically
      // unchanged, if the base_memory value has changed.
      set_req(i, new_in);
      progress = this;          // Report progress
    }
  }

  if (new_base != old_base) {
    set_req(Compile::AliasIdxBot, new_base);
    // Don't use set_base_memory(new_base), because we need to update du.
    assert(base_memory() == new_base, "");
    progress = this;
  }

  if( base_memory() == this ) {
    // a self cycle indicates this memory path is dead
    set_req(Compile::AliasIdxBot, empty_mem);
  }

  // Resolve external cycles by calling Ideal on a MergeMem base_memory
  // Recursion must occur after the self cycle check above
  if( base_memory()->is_MergeMem() ) {
    MergeMemNode *new_mbase = base_memory()->as_MergeMem();
    Node *m = phase->transform(new_mbase);  // Rollup any cycles
    if( m != NULL && (m->is_top() ||
        m->is_MergeMem() && m->as_MergeMem()->base_memory() == empty_mem) ) {
      // propagate rollup of dead cycle to self
      set_req(Compile::AliasIdxBot, empty_mem);
    }
  }

  if( base_memory() == empty_mem ) {
    progress = this;
    // Cut inputs during Parse phase only.
    // During Optimize phase a dead MergeMem node will be subsumed by Top.
    if( !can_reshape ) {
      for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
        if( in(i) != empty_mem ) { set_req(i, empty_mem); }
      }
    }
  }

  if( !progress && base_memory()->is_Phi() && can_reshape ) {
    // Check if PhiNode::Ideal's "Split phis through memory merges"
    // transform should be attempted. Look for this->phi->this cycle.
    uint merge_width = req();
    if (merge_width > Compile::AliasIdxRaw) {
      PhiNode* phi = base_memory()->as_Phi();
      for( uint i = 1; i < phi->req(); ++i ) {// For all paths in
        if (phi->in(i) == this) {
          phase->is_IterGVN()->_worklist.push(phi);
          break;
        }
      }
    }
  }

  assert(progress || verify_sparse(), "please, no dups of base");
  return progress;
}

//-------------------------set_base_memory-------------------------------------
void MergeMemNode::set_base_memory(Node *new_base) {
  Node* empty_mem = empty_memory();
  set_req(Compile::AliasIdxBot, new_base);
  assert(memory_at(req()) == new_base, "must set default memory");
  // Clear out other occurrences of new_base:
  if (new_base != empty_mem) {
    for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
      if (in(i) == new_base)  set_req(i, empty_mem);
    }
  }
}

//------------------------------out_RegMask------------------------------------
const RegMask &MergeMemNode::out_RegMask() const {
  return RegMask::Empty;
}

//------------------------------dump_spec--------------------------------------
#ifndef PRODUCT
void MergeMemNode::dump_spec(outputStream *st) const {
  st->print(" {");
  Node* base_mem = base_memory();
  for( uint i = Compile::AliasIdxRaw; i < req(); i++ ) {
    Node* mem = memory_at(i);
    if (mem == base_mem) { st->print(" -"); continue; }
    st->print( " N%d:", mem->_idx );
    Compile::current()->get_adr_type(i)->dump_on(st);
  }
  st->print(" }");
}
#endif // !PRODUCT


#ifdef ASSERT
static bool might_be_same(Node* a, Node* b) {
  if (a == b)  return true;
  if (!(a->is_Phi() || b->is_Phi()))  return false;
  // phis shift around during optimization
  return true;  // pretty stupid...
}

// verify a narrow slice (either incoming or outgoing)
static void verify_memory_slice(const MergeMemNode* m, int alias_idx, Node* n) {
  if (!VerifyAliases)       return;  // don't bother to verify unless requested
  if (is_error_reported())  return;  // muzzle asserts when debugging an error
  if (Node::in_dump())      return;  // muzzle asserts when printing
  assert(alias_idx >= Compile::AliasIdxRaw, "must not disturb base_memory or sentinel");
  assert(n != NULL, "");
  // Elide intervening MergeMem's
  while (n->is_MergeMem()) {
    n = n->as_MergeMem()->memory_at(alias_idx);
  }
  Compile* C = Compile::current();
  const TypePtr* n_adr_type = n->adr_type();
  if (n == m->empty_memory()) {
    // Implicit copy of base_memory()
  } else if (n_adr_type != TypePtr::BOTTOM) {
    assert(n_adr_type != NULL, "new memory must have a well-defined adr_type");
    assert(C->must_alias(n_adr_type, alias_idx), "new memory must match selected slice");
  } else {
    // A few places like make_runtime_call "know" that VM calls are narrow,
    // and can be used to update only the VM bits stored as TypeRawPtr::BOTTOM.
    bool expected_wide_mem = false;
    if (n == m->base_memory()) {
      expected_wide_mem = true;
    } else if (alias_idx == Compile::AliasIdxRaw ||
               n == m->memory_at(Compile::AliasIdxRaw)) {
      expected_wide_mem = true;
    } else if (!C->alias_type(alias_idx)->is_rewritable()) {
      // memory can "leak through" calls on channels that
      // are write-once.  Allow this also.
      expected_wide_mem = true;
    }
    assert(expected_wide_mem, "expected narrow slice replacement");
  }
}
#else // !ASSERT
#define verify_memory_slice(m,i,n) (0)  // PRODUCT version is no-op
#endif


//-----------------------------memory_at---------------------------------------
Node* MergeMemNode::memory_at(uint alias_idx) const {
  assert(alias_idx >= Compile::AliasIdxRaw ||
         alias_idx == Compile::AliasIdxBot && Compile::current()->AliasLevel() == 0,
         "must avoid base_memory and AliasIdxTop");

  // Otherwise, it is a narrow slice.
  Node* n = alias_idx < req() ? in(alias_idx) : empty_memory();
  Compile *C = Compile::current();
  if (is_empty_memory(n)) {
    // the array is sparse; empty slots are the "top" node
    n = base_memory();
    assert(Node::in_dump()
           || n == NULL || n->bottom_type() == Type::TOP
           || n->adr_type() == TypePtr::BOTTOM
           || n->adr_type() == TypeRawPtr::BOTTOM
           || Compile::current()->AliasLevel() == 0,
           "must be a wide memory");
    // AliasLevel == 0 if we are organizing the memory states manually.
    // See verify_memory_slice for comments on TypeRawPtr::BOTTOM.
  } else {
    // make sure the stored slice is sane
    #ifdef ASSERT
    if (is_error_reported() || Node::in_dump()) {
    } else if (might_be_same(n, base_memory())) {
      // Give it a pass:  It is a mostly harmless repetition of the base.
      // This can arise normally from node subsumption during optimization.
    } else {
      verify_memory_slice(this, alias_idx, n);
    }
    #endif
  }
  return n;
}

//---------------------------set_memory_at-------------------------------------
void MergeMemNode::set_memory_at(uint alias_idx, Node *n) {
  verify_memory_slice(this, alias_idx, n);
  Node* empty_mem = empty_memory();
  if (n == base_memory())  n = empty_mem;  // collapse default
  uint need_req = alias_idx+1;
  if (req() < need_req) {
    if (n == empty_mem)  return;  // already the default, so do not grow me
    // grow the sparse array
    do {
      add_req(empty_mem);
    } while (req() < need_req);
  }
  set_req( alias_idx, n );
}


//--------------------------iteration_setup------------------------------------
void MergeMemNode::iteration_setup(const MergeMemNode* other) {
  if (other != NULL) {
    grow_to_match(other);
    // invariant:  the finite support of mm2 is within mm->req()
    #ifdef ASSERT
    for (uint i = req(); i < other->req(); i++) {
      assert(other->is_empty_memory(other->in(i)), "slice left uncovered");
    }
    #endif
  }
  // Replace spurious copies of base_memory by top.
  Node* base_mem = base_memory();
  if (base_mem != NULL && !base_mem->is_top()) {
    for (uint i = Compile::AliasIdxBot+1, imax = req(); i < imax; i++) {
      if (in(i) == base_mem)
        set_req(i, empty_memory());
    }
  }
}

//---------------------------grow_to_match-------------------------------------
void MergeMemNode::grow_to_match(const MergeMemNode* other) {
  Node* empty_mem = empty_memory();
  assert(other->is_empty_memory(empty_mem), "consistent sentinels");
  // look for the finite support of the other memory
  for (uint i = other->req(); --i >= req(); ) {
    if (other->in(i) != empty_mem) {
      uint new_len = i+1;
      while (req() < new_len)  add_req(empty_mem);
      break;
    }
  }
}

//---------------------------verify_sparse-------------------------------------
#ifndef PRODUCT
bool MergeMemNode::verify_sparse() const {
  assert(is_empty_memory(make_empty_memory()), "sane sentinel");
  Node* base_mem = base_memory();
  // The following can happen in degenerate cases, since empty==top.
  if (is_empty_memory(base_mem))  return true;
  for (uint i = Compile::AliasIdxRaw; i < req(); i++) {
    assert(in(i) != NULL, "sane slice");
    if (in(i) == base_mem)  return false;  // should have been the sentinel value!
  }
  return true;
}

bool MergeMemStream::match_memory(Node* mem, const MergeMemNode* mm, int idx) {
  Node* n;
  n = mm->in(idx);
  if (mem == n)  return true;  // might be empty_memory()
  n = (idx == Compile::AliasIdxBot)? mm->base_memory(): mm->memory_at(idx);
  if (mem == n)  return true;
  while (n->is_Phi() && (n = n->as_Phi()->is_copy()) != NULL) {
    if (mem == n)  return true;
    if (n == NULL)  break;
  }
  return false;
}
#endif // !PRODUCT