Mercurial > hg > truffle

changeset 76:d6fe2e4959d6

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Merge
author rasbold
date Fri, 21 Mar 2008 08:32:17 -0700
parents cd0742ba123c (current diff) f68325221ce1 (diff)
children 36cd3cc4d27b d05ebaf00ed0 deb97b8ef02b
files src/cpu/x86/vm/assembler_x86_64.cpp src/share/vm/runtime/arguments.cpp src/share/vm/runtime/globals.hpp
diffstat 45 files changed, 3513 insertions(+), 758 deletions(-) [+]
line wrap: on
line diff
--- a/make/hotspot_version	Thu Mar 20 09:17:30 2008 -0500
+++ b/make/hotspot_version	Fri Mar 21 08:32:17 2008 -0700
@@ -35,7 +35,7 @@
 
 HS_MAJOR_VER=12
 HS_MINOR_VER=0
-HS_BUILD_NUMBER=01
+HS_BUILD_NUMBER=02
 
 JDK_MAJOR_VER=1
 JDK_MINOR_VER=7
--- a/src/cpu/x86/vm/assembler_x86_32.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/cpu/x86/vm/assembler_x86_32.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -2672,6 +2672,22 @@
   emit_sse_operand(dst, src);
 }
 
+void Assembler::cvtdq2pd(XMMRegister dst, XMMRegister src) {
+  assert(VM_Version::supports_sse2(), "");
+
+  emit_byte(0xF3);
+  emit_byte(0x0F);
+  emit_byte(0xE6);
+  emit_sse_operand(dst, src);
+}
+
+void Assembler::cvtdq2ps(XMMRegister dst, XMMRegister src) {
+  assert(VM_Version::supports_sse2(), "");
+
+  emit_byte(0x0F);
+  emit_byte(0x5B);
+  emit_sse_operand(dst, src);
+}
 
 emit_sse_instruction(andps,  sse,  0,    0x54, XMMRegister, XMMRegister);
 emit_sse_instruction(andpd,  sse2, 0x66, 0x54, XMMRegister, XMMRegister);
--- a/src/cpu/x86/vm/assembler_x86_32.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/cpu/x86/vm/assembler_x86_32.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -901,6 +901,8 @@
   void cvtss2sd(XMMRegister dst, XMMRegister src);
   void cvtsd2ss(XMMRegister dst, Address src);   // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value
   void cvtsd2ss(XMMRegister dst, XMMRegister src);
+  void cvtdq2pd(XMMRegister dst, XMMRegister src);
+  void cvtdq2ps(XMMRegister dst, XMMRegister src);
 
   void cvtsi2ss(XMMRegister dst, Address src);   // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value
   void cvtsi2ss(XMMRegister dst, Register src);
--- a/src/cpu/x86/vm/assembler_x86_64.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/cpu/x86/vm/assembler_x86_64.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -3372,6 +3372,21 @@
   emit_byte(0xC0 | encode);
 }
 
+void Assembler::cvtdq2pd(XMMRegister dst, XMMRegister src) {
+  emit_byte(0xF3);
+  int encode = prefix_and_encode(dst->encoding(), src->encoding());
+  emit_byte(0x0F);
+  emit_byte(0xE6);
+  emit_byte(0xC0 | encode);
+}
+
+void Assembler::cvtdq2ps(XMMRegister dst, XMMRegister src) {
+  int encode = prefix_and_encode(dst->encoding(), src->encoding());
+  emit_byte(0x0F);
+  emit_byte(0x5B);
+  emit_byte(0xC0 | encode);
+}
+
 void Assembler::cvtsd2ss(XMMRegister dst, XMMRegister src) {
   emit_byte(0xF2);
   int encode = prefix_and_encode(dst->encoding(), src->encoding());
--- a/src/cpu/x86/vm/assembler_x86_64.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/cpu/x86/vm/assembler_x86_64.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -922,6 +922,8 @@
   void cvttsd2siq(Register dst, XMMRegister src); // truncates
   void cvtss2sd(XMMRegister dst, XMMRegister src);
   void cvtsd2ss(XMMRegister dst, XMMRegister src);
+  void cvtdq2pd(XMMRegister dst, XMMRegister src);
+  void cvtdq2ps(XMMRegister dst, XMMRegister src);
 
   void pxor(XMMRegister dst, Address src);       // Xor Packed Byte Integer Values
   void pxor(XMMRegister dst, XMMRegister src);   // Xor Packed Byte Integer Values
--- a/src/cpu/x86/vm/vm_version_x86_32.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/cpu/x86/vm/vm_version_x86_32.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -321,6 +321,20 @@
         UseXmmRegToRegMoveAll = false;
       }
     }
+    if( FLAG_IS_DEFAULT(UseXmmI2F) ) {
+      if( supports_sse4a() ) {
+        UseXmmI2F = true;
+      } else {
+        UseXmmI2F = false;
+      }
+    }
+    if( FLAG_IS_DEFAULT(UseXmmI2D) ) {
+      if( supports_sse4a() ) {
+        UseXmmI2D = true;
+      } else {
+        UseXmmI2D = false;
+      }
+    }
   }
 
   if( is_intel() ) { // Intel cpus specific settings
--- a/src/cpu/x86/vm/vm_version_x86_64.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/cpu/x86/vm/vm_version_x86_64.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -265,6 +265,20 @@
         UseXmmRegToRegMoveAll = false;
       }
     }
+    if( FLAG_IS_DEFAULT(UseXmmI2F) ) {
+      if( supports_sse4a() ) {
+        UseXmmI2F = true;
+      } else {
+        UseXmmI2F = false;
+      }
+    }
+    if( FLAG_IS_DEFAULT(UseXmmI2D) ) {
+      if( supports_sse4a() ) {
+        UseXmmI2D = true;
+      } else {
+        UseXmmI2D = false;
+      }
+    }
   }
 
   if( is_intel() ) { // Intel cpus specific settings
--- a/src/cpu/x86/vm/x86_32.ad	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/cpu/x86/vm/x86_32.ad	Fri Mar 21 08:32:17 2008 -0700
@@ -10970,7 +10970,7 @@
 %}
 
 instruct convI2XD_reg(regXD dst, eRegI src) %{
-  predicate( UseSSE>=2 );
+  predicate( UseSSE>=2 && !UseXmmI2D );
   match(Set dst (ConvI2D src));
   format %{ "CVTSI2SD $dst,$src" %}
   opcode(0xF2, 0x0F, 0x2A);
@@ -10987,6 +10987,20 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct convXI2XD_reg(regXD dst, eRegI src)
+%{
+  predicate( UseSSE>=2 && UseXmmI2D );
+  match(Set dst (ConvI2D src));
+
+  format %{ "MOVD  $dst,$src\n\t"
+            "CVTDQ2PD $dst,$dst\t# i2d" %}
+  ins_encode %{
+    __ movd($dst$$XMMRegister, $src$$Register);
+    __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe(pipe_slow); // XXX
+%}
+
 instruct convI2D_mem(regD dst, memory mem) %{
   predicate( UseSSE<=1 && !Compile::current()->select_24_bit_instr());
   match(Set dst (ConvI2D (LoadI mem)));
@@ -11062,7 +11076,7 @@
 
 // Convert an int to a float in xmm; no rounding step needed.
 instruct convI2X_reg(regX dst, eRegI src) %{
-  predicate(UseSSE>=1);
+  predicate( UseSSE==1 || UseSSE>=2 && !UseXmmI2F );
   match(Set dst (ConvI2F src));
   format %{ "CVTSI2SS $dst, $src" %}
 
@@ -11071,6 +11085,20 @@
   ins_pipe( pipe_slow );
 %}
 
+ instruct convXI2X_reg(regX dst, eRegI src)
+%{
+  predicate( UseSSE>=2 && UseXmmI2F );
+  match(Set dst (ConvI2F src));
+
+  format %{ "MOVD  $dst,$src\n\t"
+            "CVTDQ2PS $dst,$dst\t# i2f" %}
+  ins_encode %{
+    __ movd($dst$$XMMRegister, $src$$Register);
+    __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe(pipe_slow); // XXX
+%}
+
 instruct convI2L_reg( eRegL dst, eRegI src, eFlagsReg cr) %{
   match(Set dst (ConvI2L src));
   effect(KILL cr);
--- a/src/cpu/x86/vm/x86_64.ad	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/cpu/x86/vm/x86_64.ad	Fri Mar 21 08:32:17 2008 -0700
@@ -10098,6 +10098,7 @@
 
 instruct convI2F_reg_reg(regF dst, rRegI src)
 %{
+  predicate(!UseXmmI2F);
   match(Set dst (ConvI2F src));
 
   format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
@@ -10118,6 +10119,7 @@
 
 instruct convI2D_reg_reg(regD dst, rRegI src)
 %{
+  predicate(!UseXmmI2D);
   match(Set dst (ConvI2D src));
 
   format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
@@ -10136,6 +10138,34 @@
   ins_pipe(pipe_slow); // XXX
 %}
 
+instruct convXI2F_reg(regF dst, rRegI src)
+%{
+  predicate(UseXmmI2F);
+  match(Set dst (ConvI2F src));
+
+  format %{ "movdl $dst, $src\n\t"
+            "cvtdq2psl $dst, $dst\t# i2f" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe(pipe_slow); // XXX
+%}
+
+instruct convXI2D_reg(regD dst, rRegI src)
+%{
+  predicate(UseXmmI2D);
+  match(Set dst (ConvI2D src));
+
+  format %{ "movdl $dst, $src\n\t"
+            "cvtdq2pdl $dst, $dst\t# i2d" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe(pipe_slow); // XXX
+%}
+
 instruct convL2F_reg_reg(regF dst, rRegL src)
 %{
   match(Set dst (ConvL2F src));
--- a/src/share/vm/includeDB_compiler2	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/includeDB_compiler2	Fri Mar 21 08:32:17 2008 -0700
@@ -164,6 +164,7 @@
 callGenerator.hpp                       type.hpp
 
 callnode.cpp                            callnode.hpp
+callnode.cpp                            bcEscapeAnalyzer.hpp
 callnode.cpp                            escape.hpp
 callnode.cpp                            locknode.hpp
 callnode.cpp                            machnode.hpp
@@ -176,7 +177,6 @@
 callnode.cpp                            runtime.hpp
 
 callnode.hpp                            connode.hpp
-callnode.hpp                            escape.hpp
 callnode.hpp                            mulnode.hpp
 callnode.hpp                            multnode.hpp
 callnode.hpp                            opcodes.hpp
@@ -347,7 +347,6 @@
 connode.cpp                             allocation.inline.hpp
 connode.cpp                             compile.hpp
 connode.cpp                             connode.hpp
-connode.cpp                             escape.hpp
 connode.cpp                             machnode.hpp
 connode.cpp                             matcher.hpp
 connode.cpp                             memnode.hpp
@@ -844,7 +843,6 @@
 phaseX.cpp                              callnode.hpp
 phaseX.cpp                              cfgnode.hpp
 phaseX.cpp                              connode.hpp
-phaseX.cpp                              escape.hpp
 phaseX.cpp                              loopnode.hpp
 phaseX.cpp                              machnode.hpp
 phaseX.cpp                              opcodes.hpp
--- a/src/share/vm/opto/c2_globals.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/c2_globals.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -382,6 +382,12 @@
   product(bool, EliminateAllocations, true,                                 \
           "Use escape analysis to eliminate allocations")                   \
                                                                             \
+  notproduct(bool, PrintEliminateAllocations, false,                        \
+          "Print out when allocations are eliminated")                      \
+                                                                            \
+  product(intx, EliminateAllocationArraySizeLimit, 64,                      \
+          "Array size (number of elements) limit for scalar replacement")   \
+                                                                            \
   product(intx, MaxLabelRootDepth, 1100,                                    \
           "Maximum times call Label_Root to prevent stack overflow")        \
 
--- a/src/share/vm/opto/callnode.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/callnode.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -230,6 +230,7 @@
   _locoff = TypeFunc::Parms;
   _stkoff = _locoff + _method->max_locals();
   _monoff = _stkoff + _method->max_stack();
+  _scloff = _monoff;
   _endoff = _monoff;
   _sp = 0;
 }
@@ -242,6 +243,7 @@
   _locoff = TypeFunc::Parms;
   _stkoff = _locoff;
   _monoff = _stkoff + stack_size;
+  _scloff = _monoff;
   _endoff = _monoff;
   _sp = 0;
 }
@@ -297,12 +299,22 @@
   return total;
 }
 
+#ifndef PRODUCT
+
 //------------------------------format_helper----------------------------------
 // Given an allocation (a Chaitin object) and a Node decide if the Node carries
 // any defined value or not.  If it does, print out the register or constant.
-#ifndef PRODUCT
-static void format_helper( PhaseRegAlloc *regalloc, outputStream* st, Node *n, const char *msg, uint i ) {
+static void format_helper( PhaseRegAlloc *regalloc, outputStream* st, Node *n, const char *msg, uint i, GrowableArray<SafePointScalarObjectNode*> *scobjs ) {
   if (n == NULL) { st->print(" NULL"); return; }
+  if (n->is_SafePointScalarObject()) {
+    // Scalar replacement.
+    SafePointScalarObjectNode* spobj = n->as_SafePointScalarObject();
+    scobjs->append_if_missing(spobj);
+    int sco_n = scobjs->find(spobj);
+    assert(sco_n >= 0, "");
+    st->print(" %s%d]=#ScObj" INT32_FORMAT, msg, i, sco_n);
+    return;
+  }
   if( OptoReg::is_valid(regalloc->get_reg_first(n))) { // Check for undefined
     char buf[50];
     regalloc->dump_register(n,buf);
@@ -342,10 +354,8 @@
     }
   }
 }
-#endif
 
 //------------------------------format-----------------------------------------
-#ifndef PRODUCT
 void JVMState::format(PhaseRegAlloc *regalloc, const Node *n, outputStream* st) const {
   st->print("        #");
   if( _method ) {
@@ -356,24 +366,25 @@
     return;
   }
   if (n->is_MachSafePoint()) {
+    GrowableArray<SafePointScalarObjectNode*> scobjs;
     MachSafePointNode *mcall = n->as_MachSafePoint();
     uint i;
     // Print locals
     for( i = 0; i < (uint)loc_size(); i++ )
-      format_helper( regalloc, st, mcall->local(this, i), "L[", i );
+      format_helper( regalloc, st, mcall->local(this, i), "L[", i, &scobjs );
     // Print stack
     for (i = 0; i < (uint)stk_size(); i++) {
       if ((uint)(_stkoff + i) >= mcall->len())
         st->print(" oob ");
       else
-       format_helper( regalloc, st, mcall->stack(this, i), "STK[", i );
+       format_helper( regalloc, st, mcall->stack(this, i), "STK[", i, &scobjs );
     }
     for (i = 0; (int)i < nof_monitors(); i++) {
       Node *box = mcall->monitor_box(this, i);
       Node *obj = mcall->monitor_obj(this, i);
       if ( OptoReg::is_valid(regalloc->get_reg_first(box)) ) {
         while( !box->is_BoxLock() )  box = box->in(1);
-        format_helper( regalloc, st, box, "MON-BOX[", i );
+        format_helper( regalloc, st, box, "MON-BOX[", i, &scobjs );
       } else {
         OptoReg::Name box_reg = BoxLockNode::stack_slot(box);
         st->print(" MON-BOX%d=%s+%d",
@@ -381,15 +392,71 @@
                    OptoReg::regname(OptoReg::c_frame_pointer),
                    regalloc->reg2offset(box_reg));
       }
-      format_helper( regalloc, st, obj, "MON-OBJ[", i );
+      format_helper( regalloc, st, obj, "MON-OBJ[", i, &scobjs );
+    }
+
+    for (i = 0; i < (uint)scobjs.length(); i++) {
+      // Scalar replaced objects.
+      st->print_cr("");
+      st->print("        # ScObj" INT32_FORMAT " ", i);
+      SafePointScalarObjectNode* spobj = scobjs.at(i);
+      ciKlass* cik = spobj->bottom_type()->is_oopptr()->klass();
+      assert(cik->is_instance_klass() ||
+             cik->is_array_klass(), "Not supported allocation.");
+      ciInstanceKlass *iklass = NULL;
+      if (cik->is_instance_klass()) {
+        cik->print_name_on(st);
+        iklass = cik->as_instance_klass();
+      } else if (cik->is_type_array_klass()) {
+        cik->as_array_klass()->base_element_type()->print_name_on(st);
+        st->print("[%d]=", spobj->n_fields());
+      } else if (cik->is_obj_array_klass()) {
+        ciType* cie = cik->as_array_klass()->base_element_type();
+        int ndim = 1;
+        while (cie->is_obj_array_klass()) {
+          ndim += 1;
+          cie = cie->as_array_klass()->base_element_type();
+        }
+        cie->print_name_on(st);
+        while (ndim-- > 0) {
+          st->print("[]");
+        }
+        st->print("[%d]=", spobj->n_fields());
+      }
+      st->print("{");
+      uint nf = spobj->n_fields();
+      if (nf > 0) {
+        uint first_ind = spobj->first_index();
+        Node* fld_node = mcall->in(first_ind);
+        ciField* cifield;
+        if (iklass != NULL) {
+          st->print(" [");
+          cifield = iklass->nonstatic_field_at(0);
+          cifield->print_name_on(st);
+          format_helper( regalloc, st, fld_node, ":", 0, &scobjs );
+        } else {
+          format_helper( regalloc, st, fld_node, "[", 0, &scobjs );
+        }
+        for (uint j = 1; j < nf; j++) {
+          fld_node = mcall->in(first_ind+j);
+          if (iklass != NULL) {
+            st->print(", [");
+            cifield = iklass->nonstatic_field_at(j);
+            cifield->print_name_on(st);
+            format_helper( regalloc, st, fld_node, ":", j, &scobjs );
+          } else {
+            format_helper( regalloc, st, fld_node, ", [", j, &scobjs );
+          }
+        }
+      }
+      st->print(" }");
     }
   }
   st->print_cr("");
   if (caller() != NULL)  caller()->format(regalloc, n, st);
 }
-#endif
 
-#ifndef PRODUCT
+
 void JVMState::dump_spec(outputStream *st) const {
   if (_method != NULL) {
     bool printed = false;
@@ -419,9 +486,8 @@
   }
   if (caller() != NULL)  caller()->dump_spec(st);
 }
-#endif
 
-#ifndef PRODUCT
+
 void JVMState::dump_on(outputStream* st) const {
   if (_map && !((uintptr_t)_map & 1)) {
     if (_map->len() > _map->req()) {  // _map->has_exceptions()
@@ -434,8 +500,8 @@
     }
     _map->dump(2);
   }
-  st->print("JVMS depth=%d loc=%d stk=%d mon=%d end=%d mondepth=%d sp=%d bci=%d method=",
-             depth(), locoff(), stkoff(), monoff(), endoff(), monitor_depth(), sp(), bci());
+  st->print("JVMS depth=%d loc=%d stk=%d mon=%d scalar=%d end=%d mondepth=%d sp=%d bci=%d method=",
+             depth(), locoff(), stkoff(), monoff(), scloff(), endoff(), monitor_depth(), sp(), bci());
   if (_method == NULL) {
     st->print_cr("(none)");
   } else {
@@ -465,6 +531,7 @@
   n->set_locoff(_locoff);
   n->set_stkoff(_stkoff);
   n->set_monoff(_monoff);
+  n->set_scloff(_scloff);
   n->set_endoff(_endoff);
   n->set_sp(_sp);
   n->set_map(_map);
@@ -557,6 +624,107 @@
   return 0;
 }
 
+//
+// Determine whether the call could modify the field of the specified
+// instance at the specified offset.
+//
+bool CallNode::may_modify(const TypePtr *addr_t, PhaseTransform *phase) {
+  const TypeOopPtr *adrInst_t  = addr_t->isa_oopptr();
+
+  // if not an InstPtr or not an instance type, assume the worst
+  if (adrInst_t == NULL || !adrInst_t->is_instance_field()) {
+    return true;
+  }
+  Compile *C = phase->C;
+  int offset = adrInst_t->offset();
+  assert(offset >= 0, "should be valid offset");
+  ciKlass* adr_k = adrInst_t->klass();
+  assert(adr_k->is_loaded() &&
+         adr_k->is_java_klass() &&
+         !adr_k->is_interface(),
+         "only non-abstract classes are expected");
+
+  int base_idx = C->get_alias_index(adrInst_t);
+  int size = BytesPerLong; // If we don't know the size, assume largest.
+  if (adrInst_t->isa_instptr()) {
+    ciField* field = C->alias_type(base_idx)->field();
+    if (field != NULL) {
+      size = field->size_in_bytes();
+    }
+  } else {
+    assert(adrInst_t->isa_aryptr(), "only arrays are expected");
+    size = type2aelembytes(adr_k->as_array_klass()->element_type()->basic_type());
+  }
+
+  ciMethod * meth = is_CallStaticJava() ?  as_CallStaticJava()->method() : NULL;
+  BCEscapeAnalyzer *bcea = (meth != NULL) ? meth->get_bcea() : NULL;
+
+  const TypeTuple * d = tf()->domain();
+  for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
+    const Type* t = d->field_at(i);
+    Node *arg = in(i);
+    const Type *at = phase->type(arg);
+    if (at == TypePtr::NULL_PTR || at == Type::TOP)
+      continue;  // null can't affect anything
+
+    const TypeOopPtr *at_ptr = at->isa_oopptr();
+    if (!arg->is_top() && (t->isa_oopptr() != NULL ||
+                           t->isa_ptr() && at_ptr != NULL)) {
+      assert(at_ptr != NULL, "expecting an OopPtr");
+      ciKlass* at_k = at_ptr->klass();
+      if ((adrInst_t->base() == at_ptr->base()) &&
+          at_k->is_loaded() &&
+          at_k->is_java_klass() &&
+          !at_k->is_interface()) {
+        // If we have found an argument matching addr_t, check if the field
+        // at the specified offset is modified.
+        int at_idx = C->get_alias_index(at_ptr->add_offset(offset)->isa_oopptr());
+        if (base_idx == at_idx &&
+            (bcea == NULL ||
+             bcea->is_arg_modified(i - TypeFunc::Parms, offset, size))) {
+          return true;
+        }
+      }
+    }
+  }
+  return false;
+}
+
+// Does this call have a direct reference to n other than debug information?
+bool CallNode::has_non_debug_use(Node *n) {
+  const TypeTuple * d = tf()->domain();
+  for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
+    Node *arg = in(i);
+    if (arg == n) {
+      return true;
+    }
+  }
+  return false;
+}
+
+// Returns the unique CheckCastPP of a call
+// or 'this' if there are several CheckCastPP
+// or returns NULL if there is no one.
+Node *CallNode::result_cast() {
+  Node *cast = NULL;
+
+  Node *p = proj_out(TypeFunc::Parms);
+  if (p == NULL)
+    return NULL;
+
+  for (DUIterator_Fast imax, i = p->fast_outs(imax); i < imax; i++) {
+    Node *use = p->fast_out(i);
+    if (use->is_CheckCastPP()) {
+      if (cast != NULL) {
+        return this;  // more than 1 CheckCastPP
+      }
+      cast = use;
+    }
+  }
+  return cast;
+}
+
+
 //=============================================================================
 uint CallJavaNode::size_of() const { return sizeof(*this); }
 uint CallJavaNode::cmp( const Node &n ) const {
@@ -765,6 +933,7 @@
 void SafePointNode::grow_stack(JVMState* jvms, uint grow_by) {
   assert((int)grow_by > 0, "sanity");
   int monoff = jvms->monoff();
+  int scloff = jvms->scloff();
   int endoff = jvms->endoff();
   assert(endoff == (int)req(), "no other states or debug info after me");
   Node* top = Compile::current()->top();
@@ -772,6 +941,7 @@
     ins_req(monoff, top);
   }
   jvms->set_monoff(monoff + grow_by);
+  jvms->set_scloff(scloff + grow_by);
   jvms->set_endoff(endoff + grow_by);
 }
 
@@ -781,6 +951,7 @@
   const int MonitorEdges = 2;
   assert(JVMState::logMonitorEdges == exact_log2(MonitorEdges), "correct MonitorEdges");
   assert(req() == jvms()->endoff(), "correct sizing");
+  int nextmon = jvms()->scloff();
   if (GenerateSynchronizationCode) {
     add_req(lock->box_node());
     add_req(lock->obj_node());
@@ -788,6 +959,7 @@
     add_req(NULL);
     add_req(NULL);
   }
+  jvms()->set_scloff(nextmon+MonitorEdges);
   jvms()->set_endoff(req());
 }
 
@@ -795,10 +967,13 @@
   // Delete last monitor from debug info
   debug_only(int num_before_pop = jvms()->nof_monitors());
   const int MonitorEdges = (1<<JVMState::logMonitorEdges);
+  int scloff = jvms()->scloff();
   int endoff = jvms()->endoff();
+  int new_scloff = scloff - MonitorEdges;
   int new_endoff = endoff - MonitorEdges;
+  jvms()->set_scloff(new_scloff);
   jvms()->set_endoff(new_endoff);
-  while (endoff > new_endoff)  del_req(--endoff);
+  while (scloff > new_scloff)  del_req(--scloff);
   assert(jvms()->nof_monitors() == num_before_pop-1, "");
 }
 
@@ -822,6 +997,63 @@
   return (TypeFunc::Parms == idx);
 }
 
+//==============  SafePointScalarObjectNode  ==============
+
+SafePointScalarObjectNode::SafePointScalarObjectNode(const TypeOopPtr* tp,
+#ifdef ASSERT
+                                                     AllocateNode* alloc,
+#endif
+                                                     uint first_index,
+                                                     uint n_fields) :
+  TypeNode(tp, 1), // 1 control input -- seems required.  Get from root.
+#ifdef ASSERT
+  _alloc(alloc),
+#endif
+  _first_index(first_index),
+  _n_fields(n_fields)
+{
+  init_class_id(Class_SafePointScalarObject);
+}
+
+
+uint SafePointScalarObjectNode::ideal_reg() const {
+  return 0; // No matching to machine instruction
+}
+
+const RegMask &SafePointScalarObjectNode::in_RegMask(uint idx) const {
+  return *(Compile::current()->matcher()->idealreg2debugmask[in(idx)->ideal_reg()]);
+}
+
+const RegMask &SafePointScalarObjectNode::out_RegMask() const {
+  return RegMask::Empty;
+}
+
+uint SafePointScalarObjectNode::match_edge(uint idx) const {
+  return 0;
+}
+
+SafePointScalarObjectNode*
+SafePointScalarObjectNode::clone(int jvms_adj, Dict* sosn_map) const {
+  void* cached = (*sosn_map)[(void*)this];
+  if (cached != NULL) {
+    return (SafePointScalarObjectNode*)cached;
+  }
+  Compile* C = Compile::current();
+  SafePointScalarObjectNode* res = (SafePointScalarObjectNode*)Node::clone();
+  res->_first_index += jvms_adj;
+  sosn_map->Insert((void*)this, (void*)res);
+  return res;
+}
+
+
+#ifndef PRODUCT
+void SafePointScalarObjectNode::dump_spec(outputStream *st) const {
+  st->print(" # fields@[%d..%d]", first_index(),
+             first_index() + n_fields() - 1);
+}
+
+#endif
+
 //=============================================================================
 uint AllocateNode::size_of() const { return sizeof(*this); }
 
@@ -1152,7 +1384,7 @@
 //=============================================================================
 Node *LockNode::Ideal(PhaseGVN *phase, bool can_reshape) {
 
-  // perform any generic optimizations first
+  // perform any generic optimizations first (returns 'this' or NULL)
   Node *result = SafePointNode::Ideal(phase, can_reshape);
 
   // Now see if we can optimize away this lock.  We don't actually
@@ -1160,7 +1392,20 @@
   // prevents macro expansion from expanding the lock.  Since we don't
   // modify the graph, the value returned from this function is the
   // one computed above.
-  if (EliminateLocks && !is_eliminated()) {
+  if (result == NULL && can_reshape && EliminateLocks && !is_eliminated()) {
+    //
+    // If we are locking an unescaped object, the lock/unlock is unnecessary
+    //
+    ConnectionGraph *cgr = Compile::current()->congraph();
+    PointsToNode::EscapeState es = PointsToNode::GlobalEscape;
+    if (cgr != NULL)
+      es = cgr->escape_state(obj_node(), phase);
+    if (es != PointsToNode::UnknownEscape && es != PointsToNode::GlobalEscape) {
+      // Mark it eliminated to update any counters
+      this->set_eliminated();
+      return result;
+    }
+
     //
     // Try lock coarsening
     //
@@ -1200,8 +1445,10 @@
           int unlocks = 0;
           for (int i = 0; i < lock_ops.length(); i++) {
             AbstractLockNode* lock = lock_ops.at(i);
-            if (lock->Opcode() == Op_Lock) locks++;
-            else                               unlocks++;
+            if (lock->Opcode() == Op_Lock)
+              locks++;
+            else
+              unlocks++;
             if (Verbose) {
               lock->dump(1);
             }
@@ -1238,7 +1485,7 @@
 //=============================================================================
 Node *UnlockNode::Ideal(PhaseGVN *phase, bool can_reshape) {
 
-  // perform any generic optimizations first
+  // perform any generic optimizations first (returns 'this' or NULL)
   Node * result = SafePointNode::Ideal(phase, can_reshape);
 
   // Now see if we can optimize away this unlock.  We don't actually
@@ -1246,66 +1493,18 @@
   // prevents macro expansion from expanding the unlock.  Since we don't
   // modify the graph, the value returned from this function is the
   // one computed above.
-  if (EliminateLocks && !is_eliminated()) {
+  // Escape state is defined after Parse phase.
+  if (result == NULL && can_reshape && EliminateLocks && !is_eliminated()) {
     //
-    // If we are unlocking an unescaped object, the lock/unlock is unnecessary
-    // We can eliminate them if there are no safepoints in the locked region.
+    // If we are unlocking an unescaped object, the lock/unlock is unnecessary.
     //
     ConnectionGraph *cgr = Compile::current()->congraph();
-    if (cgr != NULL && cgr->escape_state(obj_node(), phase) == PointsToNode::NoEscape) {
-      GrowableArray<AbstractLockNode*>   lock_ops;
-      LockNode *lock = find_matching_lock(this);
-      if (lock != NULL) {
-        lock_ops.append(this);
-        lock_ops.append(lock);
-        // find other unlocks which pair with the lock we found and add them
-        // to the list
-        Node * box = box_node();
-
-        for (DUIterator_Fast imax, i = box->fast_outs(imax); i < imax; i++) {
-          Node *use = box->fast_out(i);
-          if (use->is_Unlock() && use != this) {
-            UnlockNode *unlock1 = use->as_Unlock();
-            if (!unlock1->is_eliminated()) {
-              LockNode *lock1 = find_matching_lock(unlock1);
-              if (lock == lock1)
-                lock_ops.append(unlock1);
-              else if (lock1 == NULL) {
-               // we can't find a matching lock, we must assume the worst
-                lock_ops.trunc_to(0);
-                break;
-              }
-            }
-          }
-        }
-        if (lock_ops.length() > 0) {
-
-  #ifndef PRODUCT
-          if (PrintEliminateLocks) {
-            int locks = 0;
-            int unlocks = 0;
-            for (int i = 0; i < lock_ops.length(); i++) {
-              AbstractLockNode* lock = lock_ops.at(i);
-              if (lock->Opcode() == Op_Lock) locks++;
-              else                               unlocks++;
-              if (Verbose) {
-                lock->dump(1);
-              }
-            }
-            tty->print_cr("***Eliminated %d unescaped unlocks and %d unescaped locks", unlocks, locks);
-          }
-  #endif
-
-          // for each of the identified locks, mark them
-          // as eliminatable
-          for (int i = 0; i < lock_ops.length(); i++) {
-            AbstractLockNode* lock = lock_ops.at(i);
-
-            // Mark it eliminated to update any counters
-            lock->set_eliminated();
-          }
-        }
-      }
+    PointsToNode::EscapeState es = PointsToNode::GlobalEscape;
+    if (cgr != NULL)
+      es = cgr->escape_state(obj_node(), phase);
+    if (es != PointsToNode::UnknownEscape && es != PointsToNode::GlobalEscape) {
+      // Mark it eliminated to update any counters
+      this->set_eliminated();
     }
   }
   return result;
--- a/src/share/vm/opto/callnode.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/callnode.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -184,6 +184,7 @@
   uint              _locoff;    // Offset to locals in input edge mapping
   uint              _stkoff;    // Offset to stack in input edge mapping
   uint              _monoff;    // Offset to monitors in input edge mapping
+  uint              _scloff;    // Offset to fields of scalar objs in input edge mapping
   uint              _endoff;    // Offset to end of input edge mapping
   uint              _sp;        // Jave Expression Stack Pointer for this state
   int               _bci;       // Byte Code Index of this JVM point
@@ -207,16 +208,19 @@
   uint              stkoff() const { return _stkoff; }
   uint              argoff() const { return _stkoff + _sp; }
   uint              monoff() const { return _monoff; }
+  uint              scloff() const { return _scloff; }
   uint              endoff() const { return _endoff; }
   uint              oopoff() const { return debug_end(); }
 
   int            loc_size() const { return _stkoff - _locoff; }
   int            stk_size() const { return _monoff - _stkoff; }
-  int            mon_size() const { return _endoff - _monoff; }
+  int            mon_size() const { return _scloff - _monoff; }
+  int            scl_size() const { return _endoff - _scloff; }
 
   bool        is_loc(uint i) const { return i >= _locoff && i < _stkoff; }
   bool        is_stk(uint i) const { return i >= _stkoff && i < _monoff; }
-  bool        is_mon(uint i) const { return i >= _monoff && i < _endoff; }
+  bool        is_mon(uint i) const { return i >= _monoff && i < _scloff; }
+  bool        is_scl(uint i) const { return i >= _scloff && i < _endoff; }
 
   uint              sp()     const { return _sp; }
   int               bci()    const { return _bci; }
@@ -227,7 +231,9 @@
   uint              depth()  const { return _depth; }
   uint        debug_start()  const; // returns locoff of root caller
   uint        debug_end()    const; // returns endoff of self
-  uint        debug_size()   const { return loc_size() + sp() + mon_size(); }
+  uint        debug_size()   const {
+    return loc_size() + sp() + mon_size() + scl_size();
+  }
   uint        debug_depth()  const; // returns sum of debug_size values at all depths
 
   // Returns the JVM state at the desired depth (1 == root).
@@ -254,8 +260,11 @@
   void              set_locoff(uint off) { _locoff = off; }
   void              set_stkoff(uint off) { _stkoff = off; }
   void              set_monoff(uint off) { _monoff = off; }
+  void              set_scloff(uint off) { _scloff = off; }
   void              set_endoff(uint off) { _endoff = off; }
-  void              set_offsets(uint off) { _locoff = _stkoff = _monoff = _endoff = off; }
+  void              set_offsets(uint off) {
+    _locoff = _stkoff = _monoff = _scloff = _endoff = off;
+  }
   void              set_map(SafePointNode *map) { _map = map; }
   void              set_sp(uint sp) { _sp = sp; }
   void              set_bci(int bci) { _bci = bci; }
@@ -379,6 +388,9 @@
   void               set_next_exception(SafePointNode* n);
   bool                   has_exceptions() const { return next_exception() != NULL; }
 
+  // Does this node have a use of n other than in debug information?
+  virtual bool           has_non_debug_use(Node *n)  {return false; }
+
   // Standard Node stuff
   virtual int            Opcode() const;
   virtual bool           pinned() const { return true; }
@@ -399,6 +411,47 @@
 #endif
 };
 
+//------------------------------SafePointScalarObjectNode----------------------
+// A SafePointScalarObjectNode represents the state of a scalarized object
+// at a safepoint.
+
+class SafePointScalarObjectNode: public TypeNode {
+  uint _first_index; // First input edge index of a SafePoint node where
+                     // states of the scalarized object fields are collected.
+  uint _n_fields;    // Number of non-static fields of the scalarized object.
+  DEBUG_ONLY(AllocateNode* _alloc;)
+public:
+  SafePointScalarObjectNode(const TypeOopPtr* tp,
+#ifdef ASSERT
+                            AllocateNode* alloc,
+#endif
+                            uint first_index, uint n_fields);
+  virtual int Opcode() const;
+  virtual uint           ideal_reg() const;
+  virtual const RegMask &in_RegMask(uint) const;
+  virtual const RegMask &out_RegMask() const;
+  virtual uint           match_edge(uint idx) const;
+
+  uint first_index() const { return _first_index; }
+  uint n_fields()    const { return _n_fields; }
+  DEBUG_ONLY(AllocateNode* alloc() const { return _alloc; })
+
+  virtual uint size_of() const { return sizeof(*this); }
+
+  // Assumes that "this" is an argument to a safepoint node "s", and that
+  // "new_call" is being created to correspond to "s".  But the difference
+  // between the start index of the jvmstates of "new_call" and "s" is
+  // "jvms_adj".  Produce and return a SafePointScalarObjectNode that
+  // corresponds appropriately to "this" in "new_call".  Assumes that
+  // "sosn_map" is a map, specific to the translation of "s" to "new_call",
+  // mapping old SafePointScalarObjectNodes to new, to avoid multiple copies.
+  SafePointScalarObjectNode* clone(int jvms_adj, Dict* sosn_map) const;
+
+#ifndef PRODUCT
+  virtual void              dump_spec(outputStream *st) const;
+#endif
+};
+
 //------------------------------CallNode---------------------------------------
 // Call nodes now subsume the function of debug nodes at callsites, so they
 // contain the functionality of a full scope chain of debug nodes.
@@ -407,7 +460,6 @@
   const TypeFunc *_tf;        // Function type
   address      _entry_point;  // Address of method being called
   float        _cnt;          // Estimate of number of times called
-  PointsToNode::EscapeState _escape_state;
 
   CallNode(const TypeFunc* tf, address addr, const TypePtr* adr_type)
     : SafePointNode(tf->domain()->cnt(), NULL, adr_type),
@@ -417,7 +469,6 @@
   {
     init_class_id(Class_Call);
     init_flags(Flag_is_Call);
-    _escape_state = PointsToNode::UnknownEscape;
   }
 
   const TypeFunc* tf()        const { return _tf; }
@@ -443,6 +494,15 @@
   // the node the JVMState must be cloned.
   virtual void        clone_jvms() { }   // default is not to clone
 
+  // Returns true if the call may modify n
+  virtual bool        may_modify(const TypePtr *addr_t, PhaseTransform *phase);
+  // Does this node have a use of n other than in debug information?
+  virtual bool        has_non_debug_use(Node *n);
+  // Returns the unique CheckCastPP of a call
+  // or result projection is there are several CheckCastPP
+  // or returns NULL if there is no one.
+  Node *result_cast();
+
   virtual uint match_edge(uint idx) const;
 
 #ifndef PRODUCT
@@ -639,6 +699,9 @@
   virtual uint ideal_reg() const { return Op_RegP; }
   virtual bool        guaranteed_safepoint()  { return false; }
 
+  // allocations do not modify their arguments
+  virtual bool        may_modify(const TypePtr *addr_t, PhaseTransform *phase) { return false;}
+
   // Pattern-match a possible usage of AllocateNode.
   // Return null if no allocation is recognized.
   // The operand is the pointer produced by the (possible) allocation.
@@ -751,6 +814,9 @@
   // mark node as eliminated and update the counter if there is one
   void set_eliminated();
 
+  // locking does not modify its arguments
+  virtual bool        may_modify(const TypePtr *addr_t, PhaseTransform *phase){ return false;}
+
 #ifndef PRODUCT
   void create_lock_counter(JVMState* s);
   NamedCounter* counter() const { return _counter; }
--- a/src/share/vm/opto/cfgnode.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/cfgnode.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -704,6 +704,61 @@
   return mem;
 }
 
+//------------------------split_out_instance-----------------------------------
+// Split out an instance type from a bottom phi.
+PhiNode* PhiNode::split_out_instance(const TypePtr* at, PhaseIterGVN *igvn) const {
+  assert(type() == Type::MEMORY && (adr_type() == TypePtr::BOTTOM ||
+         adr_type() == TypeRawPtr::BOTTOM) , "bottom or raw memory required");
+
+  // Check if an appropriate node already exists.
+  Node *region = in(0);
+  for (DUIterator_Fast kmax, k = region->fast_outs(kmax); k < kmax; k++) {
+    Node* use = region->fast_out(k);
+    if( use->is_Phi()) {
+      PhiNode *phi2 = use->as_Phi();
+      if (phi2->type() == Type::MEMORY && phi2->adr_type() == at) {
+        return phi2;
+      }
+    }
+  }
+  Compile *C = igvn->C;
+  Arena *a = Thread::current()->resource_area();
+  Node_Array node_map = new Node_Array(a);
+  Node_Stack stack(a, C->unique() >> 4);
+  PhiNode *nphi = slice_memory(at);
+  igvn->register_new_node_with_optimizer( nphi );
+  node_map.map(_idx, nphi);
+  stack.push((Node *)this, 1);
+  while(!stack.is_empty()) {
+    PhiNode *ophi = stack.node()->as_Phi();
+    uint i = stack.index();
+    assert(i >= 1, "not control edge");
+    stack.pop();
+    nphi = node_map[ophi->_idx]->as_Phi();
+    for (; i < ophi->req(); i++) {
+      Node *in = ophi->in(i);
+      if (in == NULL || igvn->type(in) == Type::TOP)
+        continue;
+      Node *opt = MemNode::optimize_simple_memory_chain(in, at, igvn);
+      PhiNode *optphi = opt->is_Phi() ? opt->as_Phi() : NULL;
+      if (optphi != NULL && optphi->adr_type() == TypePtr::BOTTOM) {
+        opt = node_map[optphi->_idx];
+        if (opt == NULL) {
+          stack.push(ophi, i);
+          nphi = optphi->slice_memory(at);
+          igvn->register_new_node_with_optimizer( nphi );
+          node_map.map(optphi->_idx, nphi);
+          ophi = optphi;
+          i = 0; // will get incremented at top of loop
+          continue;
+        }
+      }
+      nphi->set_req(i, opt);
+    }
+  }
+  return nphi;
+}
+
 //------------------------verify_adr_type--------------------------------------
 #ifdef ASSERT
 void PhiNode::verify_adr_type(VectorSet& visited, const TypePtr* at) const {
@@ -1736,6 +1791,18 @@
         return result;
       }
     }
+    //
+    // Other optimizations on the memory chain
+    //
+    const TypePtr* at = adr_type();
+    for( uint i=1; i<req(); ++i ) {// For all paths in
+      Node *ii = in(i);
+      Node *new_in = MemNode::optimize_memory_chain(ii, at, phase);
+      if (ii != new_in ) {
+        set_req(i, new_in);
+        progress = this;
+      }
+    }
   }
 
   return progress;              // Return any progress
--- a/src/share/vm/opto/cfgnode.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/cfgnode.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -110,14 +110,15 @@
 // input in slot 0.
 class PhiNode : public TypeNode {
   const TypePtr* const _adr_type; // non-null only for Type::MEMORY nodes.
+  const int _inst_id;     // Instance id of the memory slice.
+  const int _inst_index;  // Alias index of the instance memory slice.
+  // Array elements references have the same alias_idx but different offset.
+  const int _inst_offset; // Offset of the instance memory slice.
   // Size is bigger to hold the _adr_type field.
   virtual uint hash() const;    // Check the type
   virtual uint cmp( const Node &n ) const;
   virtual uint size_of() const { return sizeof(*this); }
 
-  // Determine a unique non-trivial input, if any.
-  // Ignore casts if it helps.  Return NULL on failure.
-  Node* unique_input(PhaseTransform *phase);
   // Determine if CMoveNode::is_cmove_id can be used at this join point.
   Node* is_cmove_id(PhaseTransform* phase, int true_path);
 
@@ -127,8 +128,16 @@
          Input                  // Input values are [1..len)
   };
 
-  PhiNode( Node *r, const Type *t, const TypePtr* at = NULL )
-    : TypeNode(t,r->req()), _adr_type(at) {
+  PhiNode( Node *r, const Type *t, const TypePtr* at = NULL,
+           const int iid = TypeOopPtr::UNKNOWN_INSTANCE,
+           const int iidx = Compile::AliasIdxTop,
+           const int ioffs = Type::OffsetTop )
+    : TypeNode(t,r->req()),
+      _adr_type(at),
+      _inst_id(iid),
+      _inst_index(iidx),
+      _inst_offset(ioffs)
+  {
     init_class_id(Class_Phi);
     init_req(0, r);
     verify_adr_type();
@@ -139,6 +148,7 @@
   static PhiNode* make( Node* r, Node* x, const Type *t, const TypePtr* at = NULL );
   // create a new phi with narrowed memory type
   PhiNode* slice_memory(const TypePtr* adr_type) const;
+  PhiNode* split_out_instance(const TypePtr* at, PhaseIterGVN *igvn) const;
   // like make(r, x), but does not initialize the in edges to x
   static PhiNode* make_blank( Node* r, Node* x );
 
@@ -152,6 +162,10 @@
     return NULL;  // not a copy!
   }
 
+  // Determine a unique non-trivial input, if any.
+  // Ignore casts if it helps.  Return NULL on failure.
+  Node* unique_input(PhaseTransform *phase);
+
   // Check for a simple dead loop.
   enum LoopSafety { Safe = 0, Unsafe, UnsafeLoop };
   LoopSafety simple_data_loop_check(Node *in) const;
@@ -161,6 +175,18 @@
   virtual int Opcode() const;
   virtual bool pinned() const { return in(0) != 0; }
   virtual const TypePtr *adr_type() const { verify_adr_type(true); return _adr_type; }
+
+  const int inst_id()     const { return _inst_id; }
+  const int inst_index()  const { return _inst_index; }
+  const int inst_offset() const { return _inst_offset; }
+  bool is_same_inst_field(const Type* tp, int id, int index, int offset) {
+    return type()->basic_type() == tp->basic_type() &&
+           inst_id()     == id     &&
+           inst_index()  == index  &&
+           inst_offset() == offset &&
+           type()->higher_equal(tp);
+  }
+
   virtual const Type *Value( PhaseTransform *phase ) const;
   virtual Node *Identity( PhaseTransform *phase );
   virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
--- a/src/share/vm/opto/chaitin.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/chaitin.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -457,7 +457,8 @@
   bool may_be_copy_of_callee( Node *def ) const;
 
   // If nreg already contains the same constant as val then eliminate it
-  bool eliminate_copy_of_constant(Node* val, Block *current_block, Node_List& value, Node_List &regnd,
+  bool eliminate_copy_of_constant(Node* val, Node* n,
+                                  Block *current_block, Node_List& value, Node_List &regnd,
                                   OptoReg::Name nreg, OptoReg::Name nreg2);
   // Extend the node to LRG mapping
   void add_reference( const Node *node, const Node *old_node);
--- a/src/share/vm/opto/classes.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/classes.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -185,6 +185,7 @@
 macro(RoundDouble)
 macro(RoundFloat)
 macro(SafePoint)
+macro(SafePointScalarObject)
 macro(SCMemProj)
 macro(SinD)
 macro(SqrtD)
--- a/src/share/vm/opto/compile.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/compile.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -407,11 +407,6 @@
   return buf.code_size();
 }
 
-void  Compile::record_for_escape_analysis(Node* n) {
-  if (_congraph != NULL)
-    _congraph->record_for_escape_analysis(n);
-}
-
 
 // ============================================================================
 //------------------------------Compile standard-------------------------------
@@ -494,9 +489,6 @@
   PhaseGVN gvn(node_arena(), estimated_size);
   set_initial_gvn(&gvn);
 
-  if (_do_escape_analysis)
-    _congraph = new ConnectionGraph(this);
-
   { // Scope for timing the parser
     TracePhase t3("parse", &_t_parser, true);
 
@@ -581,6 +573,8 @@
   NOT_PRODUCT( verify_graph_edges(); )
 
   // Perform escape analysis
+  if (_do_escape_analysis)
+    _congraph = new ConnectionGraph(this);
   if (_congraph != NULL) {
     NOT_PRODUCT( TracePhase t2("escapeAnalysis", &_t_escapeAnalysis, TimeCompiler); )
     _congraph->compute_escape();
--- a/src/share/vm/opto/compile.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/compile.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -485,7 +485,6 @@
   PhaseGVN*         initial_gvn()               { return _initial_gvn; }
   Unique_Node_List* for_igvn()                  { return _for_igvn; }
   inline void       record_for_igvn(Node* n);   // Body is after class Unique_Node_List.
-  void              record_for_escape_analysis(Node* n);
   void          set_initial_gvn(PhaseGVN *gvn)           { _initial_gvn = gvn; }
   void          set_for_igvn(Unique_Node_List *for_igvn) { _for_igvn = for_igvn; }
 
@@ -606,8 +605,20 @@
 
   // Build OopMaps for each GC point
   void BuildOopMaps();
-  // Append debug info for the node to the array
-  void FillLocArray( int idx, Node *local, GrowableArray<ScopeValue*> *array );
+
+  // Append debug info for the node "local" at safepoint node "sfpt" to the
+  // "array",   May also consult and add to "objs", which describes the
+  // scalar-replaced objects.
+  void FillLocArray( int idx, MachSafePointNode* sfpt,
+                     Node *local, GrowableArray<ScopeValue*> *array,
+                     GrowableArray<ScopeValue*> *objs );
+
+  // If "objs" contains an ObjectValue whose id is "id", returns it, else NULL.
+  static ObjectValue* sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id);
+  // Requres that "objs" does not contains an ObjectValue whose id matches
+  // that of "sv.  Appends "sv".
+  static void set_sv_for_object_node(GrowableArray<ScopeValue*> *objs,
+                                     ObjectValue* sv );
 
   // Process an OopMap Element while emitting nodes
   void Process_OopMap_Node(MachNode *mach, int code_offset);
--- a/src/share/vm/opto/doCall.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/doCall.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -390,6 +390,8 @@
   }
 
   if (cg->is_inline()) {
+    // Accumulate has_loops estimate
+    C->set_has_loops(C->has_loops() || call_method->has_loops());
     C->env()->notice_inlined_method(call_method);
   }
 
--- a/src/share/vm/opto/escape.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/escape.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -60,9 +60,9 @@
 
 static char *esc_names[] = {
   "UnknownEscape",
-  "NoEscape     ",
-  "ArgEscape    ",
-  "GlobalEscape "
+  "NoEscape",
+  "ArgEscape",
+  "GlobalEscape"
 };
 
 static char *edge_type_suffix[] = {
@@ -75,7 +75,7 @@
 void PointsToNode::dump() const {
   NodeType nt = node_type();
   EscapeState es = escape_state();
-  tty->print("%s  %s  [[", node_type_names[(int) nt], esc_names[(int) es]);
+  tty->print("%s %s %s [[", node_type_names[(int) nt], esc_names[(int) es], _scalar_replaceable ? "" : "NSR");
   for (uint i = 0; i < edge_count(); i++) {
     tty->print(" %d%s", edge_target(i), edge_type_suffix[(int) edge_type(i)]);
   }
@@ -91,9 +91,11 @@
   _collecting = true;
   this->_compile = C;
   const PointsToNode &dummy = PointsToNode();
-  _nodes = new(C->comp_arena()) GrowableArray<PointsToNode>(C->comp_arena(), (int) INITIAL_NODE_COUNT, 0, dummy);
+  int sz = C->unique();
+  _nodes = new(C->comp_arena()) GrowableArray<PointsToNode>(C->comp_arena(), sz, sz, dummy);
   _phantom_object = C->top()->_idx;
   PointsToNode *phn = ptnode_adr(_phantom_object);
+  phn->_node = C->top();
   phn->set_node_type(PointsToNode::JavaObject);
   phn->set_escape_state(PointsToNode::GlobalEscape);
 }
@@ -121,8 +123,20 @@
     f->add_edge(to_i, PointsToNode::DeferredEdge);
 }
 
-int ConnectionGraph::type_to_offset(const Type *t) {
-  const TypePtr *t_ptr = t->isa_ptr();
+int ConnectionGraph::address_offset(Node* adr, PhaseTransform *phase) {
+  const Type *adr_type = phase->type(adr);
+  if (adr->is_AddP() && adr_type->isa_oopptr() == NULL &&
+      adr->in(AddPNode::Address)->is_Proj() &&
+      adr->in(AddPNode::Address)->in(0)->is_Allocate()) {
+    // We are computing a raw address for a store captured by an Initialize
+    // compute an appropriate address type. AddP cases #3 and #5 (see below).
+    int offs = (int)phase->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
+    assert(offs != Type::OffsetBot ||
+           adr->in(AddPNode::Address)->in(0)->is_AllocateArray(),
+           "offset must be a constant or it is initialization of array");
+    return offs;
+  }
+  const TypePtr *t_ptr = adr_type->isa_ptr();
   assert(t_ptr != NULL, "must be a pointer type");
   return t_ptr->offset();
 }
@@ -147,12 +161,28 @@
     npt->set_escape_state(es);
 }
 
+void ConnectionGraph::add_node(Node *n, PointsToNode::NodeType nt,
+                               PointsToNode::EscapeState es, bool done) {
+  PointsToNode* ptadr = ptnode_adr(n->_idx);
+  ptadr->_node = n;
+  ptadr->set_node_type(nt);
+
+  // inline set_escape_state(idx, es);
+  PointsToNode::EscapeState old_es = ptadr->escape_state();
+  if (es > old_es)
+    ptadr->set_escape_state(es);
+
+  if (done)
+    _processed.set(n->_idx);
+}
+
 PointsToNode::EscapeState ConnectionGraph::escape_state(Node *n, PhaseTransform *phase) {
   uint idx = n->_idx;
   PointsToNode::EscapeState es;
 
-  // If we are still collecting we don't know the answer yet
-  if (_collecting)
+  // If we are still collecting or there were no non-escaping allocations
+  // we don't know the answer yet
+  if (_collecting || !_has_allocations)
     return PointsToNode::UnknownEscape;
 
   // if the node was created after the escape computation, return
@@ -169,9 +199,9 @@
   // compute max escape state of anything this node could point to
   VectorSet ptset(Thread::current()->resource_area());
   PointsTo(ptset, n, phase);
-  for( VectorSetI i(&ptset); i.test() && es != PointsToNode::GlobalEscape; ++i ) {
+  for(VectorSetI i(&ptset); i.test() && es != PointsToNode::GlobalEscape; ++i) {
     uint pt = i.elem;
-    PointsToNode::EscapeState pes = _nodes->at(pt).escape_state();
+    PointsToNode::EscapeState pes = _nodes->adr_at(pt)->escape_state();
     if (pes > es)
       es = pes;
   }
@@ -185,7 +215,7 @@
   VectorSet visited(Thread::current()->resource_area());
   GrowableArray<uint>  worklist;
 
-  n = skip_casts(n);
+  n = n->uncast();
   PointsToNode  npt = _nodes->at_grow(n->_idx);
 
   // If we have a JavaObject, return just that object
@@ -193,39 +223,33 @@
     ptset.set(n->_idx);
     return;
   }
-  // we may have a Phi which has not been processed
-  if (npt._node == NULL) {
-    assert(n->is_Phi(), "unprocessed node must be a Phi");
-    record_for_escape_analysis(n);
-    npt = _nodes->at(n->_idx);
-  }
+  assert(npt._node != NULL, "unregistered node");
+
   worklist.push(n->_idx);
   while(worklist.length() > 0) {
     int ni = worklist.pop();
     PointsToNode pn = _nodes->at_grow(ni);
-    if (!visited.test(ni)) {
-      visited.set(ni);
-
+    if (!visited.test_set(ni)) {
       // ensure that all inputs of a Phi have been processed
-      if (_collecting && pn._node->is_Phi()) {
-        PhiNode *phi = pn._node->as_Phi();
-        process_phi_escape(phi, phase);
-      }
+      assert(!_collecting || !pn._node->is_Phi() || _processed.test(ni),"");
 
       int edges_processed = 0;
       for (uint e = 0; e < pn.edge_count(); e++) {
+        uint etgt = pn.edge_target(e);
         PointsToNode::EdgeType et = pn.edge_type(e);
         if (et == PointsToNode::PointsToEdge) {
-          ptset.set(pn.edge_target(e));
+          ptset.set(etgt);
           edges_processed++;
         } else if (et == PointsToNode::DeferredEdge) {
-          worklist.push(pn.edge_target(e));
+          worklist.push(etgt);
           edges_processed++;
+        } else {
+          assert(false,"neither PointsToEdge or DeferredEdge");
         }
       }
       if (edges_processed == 0) {
-        // no deferred or pointsto edges found.  Assume the value was set outside
-        // this method.  Add the phantom object to the pointsto set.
+        // no deferred or pointsto edges found.  Assume the value was set
+        // outside this method.  Add the phantom object to the pointsto set.
         ptset.set(_phantom_object);
       }
     }
@@ -239,20 +263,23 @@
   PointsToNode *ptn = ptnode_adr(ni);
 
   while(i < ptn->edge_count()) {
+    uint t = ptn->edge_target(i);
+    PointsToNode *ptt = ptnode_adr(t);
     if (ptn->edge_type(i) != PointsToNode::DeferredEdge) {
       i++;
     } else {
-      uint t = ptn->edge_target(i);
-      PointsToNode *ptt = ptnode_adr(t);
       ptn->remove_edge(t, PointsToNode::DeferredEdge);
-      if(!visited.test(t)) {
-        visited.set(t);
+      if(!visited.test_set(t)) {
         for (uint j = 0; j < ptt->edge_count(); j++) {
           uint n1 = ptt->edge_target(j);
           PointsToNode *pt1 = ptnode_adr(n1);
           switch(ptt->edge_type(j)) {
             case PointsToNode::PointsToEdge:
-               add_pointsto_edge(ni, n1);
+              add_pointsto_edge(ni, n1);
+              if(n1 == _phantom_object) {
+                // Special case - field set outside (globally escaping).
+                ptn->set_escape_state(PointsToNode::GlobalEscape);
+              }
               break;
             case PointsToNode::DeferredEdge:
               add_deferred_edge(ni, n1);
@@ -291,8 +318,8 @@
   }
 }
 
-//  Add a deferred  edge from node given by "from_i" to any field of adr_i whose offset
-//  matches "offset"
+// Add a deferred  edge from node given by "from_i" to any field of adr_i
+// whose offset matches "offset".
 void ConnectionGraph::add_deferred_edge_to_fields(uint from_i, uint adr_i, int offs) {
   PointsToNode an = _nodes->at_grow(adr_i);
   for (uint fe = 0; fe < an.edge_count(); fe++) {
@@ -310,25 +337,108 @@
   }
 }
 
-//
-// Search memory chain of "mem" to find a MemNode whose address
-// is the specified alias index.  Returns the MemNode found or the
-// first non-MemNode encountered.
-//
-Node *ConnectionGraph::find_mem(Node *mem, int alias_idx, PhaseGVN  *igvn) {
-  if (mem == NULL)
-    return mem;
-  while (mem->is_Mem()) {
-    const Type *at = igvn->type(mem->in(MemNode::Address));
-    if (at != Type::TOP) {
-      assert (at->isa_ptr() != NULL, "pointer type required.");
-      int idx = _compile->get_alias_index(at->is_ptr());
-      if (idx == alias_idx)
-        break;
-    }
-    mem = mem->in(MemNode::Memory);
+// Helper functions
+
+static Node* get_addp_base(Node *addp) {
+  assert(addp->is_AddP(), "must be AddP");
+  //
+  // AddP cases for Base and Address inputs:
+  // case #1. Direct object's field reference:
+  //     Allocate
+  //       |
+  //     Proj #5 ( oop result )
+  //       |
+  //     CheckCastPP (cast to instance type)
+  //      | |
+  //     AddP  ( base == address )
+  //
+  // case #2. Indirect object's field reference:
+  //      Phi
+  //       |
+  //     CastPP (cast to instance type)
+  //      | |
+  //     AddP  ( base == address )
+  //
+  // case #3. Raw object's field reference for Initialize node:
+  //      Allocate
+  //        |
+  //      Proj #5 ( oop result )
+  //  top   |
+  //     \  |
+  //     AddP  ( base == top )
+  //
+  // case #4. Array's element reference:
+  //   {CheckCastPP | CastPP}
+  //     |  | |
+  //     |  AddP ( array's element offset )
+  //     |  |
+  //     AddP ( array's offset )
+  //
+  // case #5. Raw object's field reference for arraycopy stub call:
+  //          The inline_native_clone() case when the arraycopy stub is called
+  //          after the allocation before Initialize and CheckCastPP nodes.
+  //      Allocate
+  //        |
+  //      Proj #5 ( oop result )
+  //       | |
+  //       AddP  ( base == address )
+  //
+  // case #6. Constant Pool or ThreadLocal or Raw object's field reference:
+  //      ConP # Object from Constant Pool.
+  //  top   |
+  //     \  |
+  //     AddP  ( base == top )
+  //
+  Node *base = addp->in(AddPNode::Base)->uncast();
+  if (base->is_top()) { // The AddP case #3 and #6.
+    base = addp->in(AddPNode::Address)->uncast();
+    assert(base->Opcode() == Op_ConP || base->Opcode() == Op_ThreadLocal ||
+           base->is_Mem() && base->bottom_type() == TypeRawPtr::NOTNULL ||
+           base->is_Proj() && base->in(0)->is_Allocate(), "sanity");
   }
-  return mem;
+  return base;
+}
+
+static Node* find_second_addp(Node* addp, Node* n) {
+  assert(addp->is_AddP() && addp->outcnt() > 0, "Don't process dead nodes");
+
+  Node* addp2 = addp->raw_out(0);
+  if (addp->outcnt() == 1 && addp2->is_AddP() &&
+      addp2->in(AddPNode::Base) == n &&
+      addp2->in(AddPNode::Address) == addp) {
+
+    assert(addp->in(AddPNode::Base) == n, "expecting the same base");
+    //
+    // Find array's offset to push it on worklist first and
+    // as result process an array's element offset first (pushed second)
+    // to avoid CastPP for the array's offset.
+    // Otherwise the inserted CastPP (LocalVar) will point to what
+    // the AddP (Field) points to. Which would be wrong since
+    // the algorithm expects the CastPP has the same point as
+    // as AddP's base CheckCastPP (LocalVar).
+    //
+    //    ArrayAllocation
+    //     |
+    //    CheckCastPP
+    //     |
+    //    memProj (from ArrayAllocation CheckCastPP)
+    //     |  ||
+    //     |  ||   Int (element index)
+    //     |  ||    |   ConI (log(element size))
+    //     |  ||    |   /
+    //     |  ||   LShift
+    //     |  ||  /
+    //     |  AddP (array's element offset)
+    //     |  |
+    //     |  | ConI (array's offset: #12(32-bits) or #24(64-bits))
+    //     | / /
+    //     AddP (array's offset)
+    //      |
+    //     Load/Store (memory operation on array's element)
+    //
+    return addp2;
+  }
+  return NULL;
 }
 
 //
@@ -336,24 +446,33 @@
 // address of a field of an instance
 //
 void ConnectionGraph::split_AddP(Node *addp, Node *base,  PhaseGVN  *igvn) {
+  const TypeOopPtr *base_t = igvn->type(base)->isa_oopptr();
+  assert(base_t != NULL && base_t->is_instance(), "expecting instance oopptr");
   const TypeOopPtr *t = igvn->type(addp)->isa_oopptr();
-  const TypeOopPtr *base_t = igvn->type(base)->isa_oopptr();
-  assert(t != NULL,  "expecting oopptr");
-  assert(base_t != NULL && base_t->is_instance(), "expecting instance oopptr");
+  if (t == NULL) {
+    // We are computing a raw address for a store captured by an Initialize
+    // compute an appropriate address type.
+    assert(igvn->type(addp) == TypeRawPtr::NOTNULL, "must be raw pointer");
+    assert(addp->in(AddPNode::Address)->is_Proj(), "base of raw address must be result projection from allocation");
+    int offs = (int)igvn->find_intptr_t_con(addp->in(AddPNode::Offset), Type::OffsetBot);
+    assert(offs != Type::OffsetBot, "offset must be a constant");
+    t = base_t->add_offset(offs)->is_oopptr();
+  }
   uint inst_id =  base_t->instance_id();
   assert(!t->is_instance() || t->instance_id() == inst_id,
                              "old type must be non-instance or match new type");
   const TypeOopPtr *tinst = base_t->add_offset(t->offset())->is_oopptr();
-  // ensure an alias index is allocated for the instance type
+  // Do NOT remove the next call: ensure an new alias index is allocated
+  // for the instance type
   int alias_idx = _compile->get_alias_index(tinst);
   igvn->set_type(addp, tinst);
   // record the allocation in the node map
   set_map(addp->_idx, get_map(base->_idx));
-  // if the Address input is not the appropriate instance type (due to intervening
-  // casts,) insert a cast
+  // if the Address input is not the appropriate instance type
+  // (due to intervening casts,) insert a cast
   Node *adr = addp->in(AddPNode::Address);
   const TypeOopPtr  *atype = igvn->type(adr)->isa_oopptr();
-  if (atype->instance_id() != inst_id) {
+  if (atype != NULL && atype->instance_id() != inst_id) {
     assert(!atype->is_instance(), "no conflicting instances");
     const TypeOopPtr *new_atype = base_t->add_offset(atype->offset())->isa_oopptr();
     Node *acast = new (_compile, 2) CastPPNode(adr, new_atype);
@@ -372,8 +491,9 @@
     addp->set_req(AddPNode::Base, bcast);
     addp->set_req(AddPNode::Address, acast);
     igvn->hash_insert(addp);
-    record_for_optimizer(addp);
   }
+  // Put on IGVN worklist since at least addp's type was changed above.
+  record_for_optimizer(addp);
 }
 
 //
@@ -386,12 +506,11 @@
   new_created = false;
   int phi_alias_idx = C->get_alias_index(orig_phi->adr_type());
   // nothing to do if orig_phi is bottom memory or matches alias_idx
-  if (phi_alias_idx == Compile::AliasIdxBot || phi_alias_idx == alias_idx) {
+  if (phi_alias_idx == alias_idx) {
     return orig_phi;
   }
   // have we already created a Phi for this alias index?
   PhiNode *result = get_map_phi(orig_phi->_idx);
-  const TypePtr *atype = C->get_adr_type(alias_idx);
   if (result != NULL && C->get_alias_index(result->adr_type()) == alias_idx) {
     return result;
   }
@@ -404,8 +523,8 @@
     }
     return NULL;
   }
-
   orig_phi_worklist.append_if_missing(orig_phi);
+  const TypePtr *atype = C->get_adr_type(alias_idx);
   result = PhiNode::make(orig_phi->in(0), NULL, Type::MEMORY, atype);
   set_map_phi(orig_phi->_idx, result);
   igvn->set_type(result, result->bottom_type());
@@ -423,7 +542,7 @@
   assert(alias_idx != Compile::AliasIdxBot, "can't split out bottom memory");
   Compile *C = _compile;
   bool new_phi_created;
-  PhiNode *result =  create_split_phi(orig_phi, alias_idx, orig_phi_worklist, igvn, new_phi_created);
+  PhiNode *result = create_split_phi(orig_phi, alias_idx, orig_phi_worklist, igvn, new_phi_created);
   if (!new_phi_created) {
     return result;
   }
@@ -436,20 +555,20 @@
   bool finished = false;
   while(!finished) {
     while (idx < phi->req()) {
-      Node *mem = find_mem(phi->in(idx), alias_idx, igvn);
+      Node *mem = find_inst_mem(phi->in(idx), alias_idx, orig_phi_worklist, igvn);
       if (mem != NULL && mem->is_Phi()) {
-        PhiNode *nphi = create_split_phi(mem->as_Phi(), alias_idx, orig_phi_worklist, igvn, new_phi_created);
+        PhiNode *newphi = create_split_phi(mem->as_Phi(), alias_idx, orig_phi_worklist, igvn, new_phi_created);
         if (new_phi_created) {
           // found an phi for which we created a new split, push current one on worklist and begin
           // processing new one
           phi_list.push(phi);
           cur_input.push(idx);
           phi = mem->as_Phi();
-          result = nphi;
+          result = newphi;
           idx = 1;
           continue;
         } else {
-          mem = nphi;
+          mem = newphi;
         }
       }
       if (C->failing()) {
@@ -461,23 +580,124 @@
     // verify that the new Phi has an input for each input of the original
     assert( phi->req() == result->req(), "must have same number of inputs.");
     assert( result->in(0) != NULL && result->in(0) == phi->in(0), "regions must match");
+#endif
+    // Check if all new phi's inputs have specified alias index.
+    // Otherwise use old phi.
     for (uint i = 1; i < phi->req(); i++) {
-      assert((phi->in(i) == NULL) == (result->in(i) == NULL), "inputs must correspond.");
+      Node* in = result->in(i);
+      assert((phi->in(i) == NULL) == (in == NULL), "inputs must correspond.");
     }
-#endif
     // we have finished processing a Phi, see if there are any more to do
     finished = (phi_list.length() == 0 );
     if (!finished) {
       phi = phi_list.pop();
       idx = cur_input.pop();
-      PhiNode *prev_phi = get_map_phi(phi->_idx);
-      prev_phi->set_req(idx++, result);
-      result = prev_phi;
+      PhiNode *prev_result = get_map_phi(phi->_idx);
+      prev_result->set_req(idx++, result);
+      result = prev_result;
     }
   }
   return result;
 }
 
+
+//
+// The next methods are derived from methods in MemNode.
+//
+static Node *step_through_mergemem(MergeMemNode *mmem, int alias_idx, const TypeOopPtr *tinst) {
+  Node *mem = mmem;
+  // 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.
+  if( tinst->base() != Type::AnyPtr &&
+      !(tinst->klass()->is_java_lang_Object() &&
+        tinst->offset() == Type::OffsetBot) ) {
+    mem = mmem->memory_at(alias_idx);
+    // Update input if it is progress over what we have now
+  }
+  return mem;
+}
+
+//
+// Search memory chain of "mem" to find a MemNode whose address
+// is the specified alias index.
+//
+Node* ConnectionGraph::find_inst_mem(Node *orig_mem, int alias_idx, GrowableArray<PhiNode *>  &orig_phis, PhaseGVN *phase) {
+  if (orig_mem == NULL)
+    return orig_mem;
+  Compile* C = phase->C;
+  const TypeOopPtr *tinst = C->get_adr_type(alias_idx)->isa_oopptr();
+  bool is_instance = (tinst != NULL) && tinst->is_instance();
+  Node *prev = NULL;
+  Node *result = orig_mem;
+  while (prev != result) {
+    prev = result;
+    if (result->is_Mem()) {
+      MemNode *mem = result->as_Mem();
+      const Type *at = phase->type(mem->in(MemNode::Address));
+      if (at != Type::TOP) {
+        assert (at->isa_ptr() != NULL, "pointer type required.");
+        int idx = C->get_alias_index(at->is_ptr());
+        if (idx == alias_idx)
+          break;
+      }
+      result = mem->in(MemNode::Memory);
+    }
+    if (!is_instance)
+      continue;  // don't search further for non-instance types
+    // 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(tinst, 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 != tinst->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()) {
+      MergeMemNode *mmem = result->as_MergeMem();
+      result = step_through_mergemem(mmem, alias_idx, tinst);
+      if (result == mmem->base_memory()) {
+        // Didn't find instance memory, search through general slice recursively.
+        result = mmem->memory_at(C->get_general_index(alias_idx));
+        result = find_inst_mem(result, alias_idx, orig_phis, phase);
+        if (C->failing()) {
+          return NULL;
+        }
+        mmem->set_memory_at(alias_idx, result);
+      }
+    } else if (result->is_Phi() &&
+               C->get_alias_index(result->as_Phi()->adr_type()) != alias_idx) {
+      Node *un = result->as_Phi()->unique_input(phase);
+      if (un != NULL) {
+        result = un;
+      } else {
+        break;
+      }
+    }
+  }
+  if (is_instance && result->is_Phi()) {
+    PhiNode *mphi = result->as_Phi();
+    assert(mphi->bottom_type() == Type::MEMORY, "memory phi required");
+    const TypePtr *t = mphi->adr_type();
+    if (C->get_alias_index(t) != alias_idx) {
+      result = split_memory_phi(mphi, alias_idx, orig_phis, phase);
+    }
+  }
+  // the result is either MemNode, PhiNode, InitializeNode.
+  return result;
+}
+
+
 //
 //  Convert the types of unescaped object to instance types where possible,
 //  propagate the new type information through the graph, and update memory
@@ -576,56 +796,101 @@
   VectorSet visited(Thread::current()->resource_area());
   VectorSet ptset(Thread::current()->resource_area());
 
-  //  Phase 1:  Process possible allocations from alloc_worklist.  Create instance
-  //            types for the CheckCastPP for allocations where possible.
+
+  //  Phase 1:  Process possible allocations from alloc_worklist.
+  //  Create instance types for the CheckCastPP for allocations where possible.
   while (alloc_worklist.length() != 0) {
     Node *n = alloc_worklist.pop();
     uint ni = n->_idx;
+    const TypeOopPtr* tinst = NULL;
     if (n->is_Call()) {
       CallNode *alloc = n->as_Call();
       // copy escape information to call node
-      PointsToNode ptn = _nodes->at(alloc->_idx);
+      PointsToNode* ptn = _nodes->adr_at(alloc->_idx);
       PointsToNode::EscapeState es = escape_state(alloc, igvn);
-      alloc->_escape_state = es;
-      // find CheckCastPP of call return value
-      n = alloc->proj_out(TypeFunc::Parms);
-      if (n != NULL && n->outcnt() == 1) {
-        n = n->unique_out();
-        if (n->Opcode() != Op_CheckCastPP) {
-          continue;
-        }
-      } else {
+      // We have an allocation or call which returns a Java object,
+      // see if it is unescaped.
+      if (es != PointsToNode::NoEscape || !ptn->_scalar_replaceable)
         continue;
-      }
-      // we have an allocation or call which returns a Java object, see if it is unescaped
-      if (es != PointsToNode::NoEscape || !ptn._unique_type) {
-        continue; //  can't make a unique type
-      }
       if (alloc->is_Allocate()) {
         // Set the scalar_replaceable flag before the next check.
         alloc->as_Allocate()->_is_scalar_replaceable = true;
       }
-
+      // find CheckCastPP of call return value
+      n = alloc->result_cast();
+      if (n == NULL ||          // No uses accept Initialize or
+          !n->is_CheckCastPP()) // not unique CheckCastPP.
+        continue;
+      // The inline code for Object.clone() casts the allocation result to
+      // java.lang.Object and then to the the actual type of the allocated
+      // object. Detect this case and use the second cast.
+      if (alloc->is_Allocate() && n->as_Type()->type() == TypeInstPtr::NOTNULL
+          && igvn->type(alloc->in(AllocateNode::KlassNode)) != TypeKlassPtr::OBJECT) {
+        Node *cast2 = NULL;
+        for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
+          Node *use = n->fast_out(i);
+          if (use->is_CheckCastPP()) {
+            cast2 = use;
+            break;
+          }
+        }
+        if (cast2 != NULL) {
+          n = cast2;
+        } else {
+          continue;
+        }
+      }
+      set_escape_state(n->_idx, es);
+      // in order for an object to be stackallocatable, it must be:
+      //   - a direct allocation (not a call returning an object)
+      //   - non-escaping
+      //   - eligible to be a unique type
+      //   - not determined to be ineligible by escape analysis
       set_map(alloc->_idx, n);
       set_map(n->_idx, alloc);
-      const TypeInstPtr *t = igvn->type(n)->isa_instptr();
-      // Unique types which are arrays are not currently supported.
-      // The check for AllocateArray is needed in case an array
-      // allocation is immediately cast to Object
-      if (t == NULL || alloc->is_AllocateArray())
+      const TypeOopPtr *t = igvn->type(n)->isa_oopptr();
+      if (t == NULL)
         continue;  // not a TypeInstPtr
-      const TypeOopPtr *tinst = t->cast_to_instance(ni);
+      tinst = t->cast_to_instance(ni);
       igvn->hash_delete(n);
       igvn->set_type(n,  tinst);
       n->raise_bottom_type(tinst);
       igvn->hash_insert(n);
+      record_for_optimizer(n);
+      if (alloc->is_Allocate() && ptn->_scalar_replaceable &&
+          (t->isa_instptr() || t->isa_aryptr())) {
+        // An allocation may have an Initialize which has raw stores. Scan
+        // the users of the raw allocation result and push AddP users
+        // on alloc_worklist.
+        Node *raw_result = alloc->proj_out(TypeFunc::Parms);
+        assert (raw_result != NULL, "must have an allocation result");
+        for (DUIterator_Fast imax, i = raw_result->fast_outs(imax); i < imax; i++) {
+          Node *use = raw_result->fast_out(i);
+          if (use->is_AddP() && use->outcnt() > 0) { // Don't process dead nodes
+            Node* addp2 = find_second_addp(use, raw_result);
+            if (addp2 != NULL) {
+              assert(alloc->is_AllocateArray(),"array allocation was expected");
+              alloc_worklist.append_if_missing(addp2);
+            }
+            alloc_worklist.append_if_missing(use);
+          } else if (use->is_Initialize()) {
+            memnode_worklist.append_if_missing(use);
+          }
+        }
+      }
     } else if (n->is_AddP()) {
       ptset.Clear();
-      PointsTo(ptset, n->in(AddPNode::Address), igvn);
+      PointsTo(ptset, get_addp_base(n), igvn);
       assert(ptset.Size() == 1, "AddP address is unique");
-      Node *base = get_map(ptset.getelem());
+      uint elem = ptset.getelem(); // Allocation node's index
+      if (elem == _phantom_object)
+        continue; // Assume the value was set outside this method.
+      Node *base = get_map(elem);  // CheckCastPP node
       split_AddP(n, base, igvn);
-    } else if (n->is_Phi() || n->Opcode() == Op_CastPP || n->Opcode() == Op_CheckCastPP) {
+      tinst = igvn->type(base)->isa_oopptr();
+    } else if (n->is_Phi() ||
+               n->is_CheckCastPP() ||
+               (n->is_ConstraintCast() && n->Opcode() == Op_CastPP)) {
       if (visited.test_set(n->_idx)) {
         assert(n->is_Phi(), "loops only through Phi's");
         continue;  // already processed
@@ -633,17 +898,23 @@
       ptset.Clear();
       PointsTo(ptset, n, igvn);
       if (ptset.Size() == 1) {
+        uint elem = ptset.getelem(); // Allocation node's index
+        if (elem == _phantom_object)
+          continue; // Assume the value was set outside this method.
+        Node *val = get_map(elem);   // CheckCastPP node
         TypeNode *tn = n->as_Type();
-        Node *val = get_map(ptset.getelem());
-        const TypeInstPtr *val_t = igvn->type(val)->isa_instptr();;
-        assert(val_t != NULL && val_t->is_instance(), "instance type expected.");
-        const TypeInstPtr *tn_t = igvn->type(tn)->isa_instptr();;
+        tinst = igvn->type(val)->isa_oopptr();
+        assert(tinst != NULL && tinst->is_instance() &&
+               tinst->instance_id() == elem , "instance type expected.");
+        const TypeOopPtr *tn_t = igvn->type(tn)->isa_oopptr();
 
-        if (tn_t != NULL && val_t->cast_to_instance(TypeOopPtr::UNKNOWN_INSTANCE)->higher_equal(tn_t)) {
+        if (tn_t != NULL &&
+ tinst->cast_to_instance(TypeOopPtr::UNKNOWN_INSTANCE)->higher_equal(tn_t)) {
           igvn->hash_delete(tn);
-          igvn->set_type(tn, val_t);
-          tn->set_type(val_t);
+          igvn->set_type(tn, tinst);
+          tn->set_type(tinst);
           igvn->hash_insert(tn);
+          record_for_optimizer(n);
         }
       }
     } else {
@@ -653,13 +924,38 @@
     for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
       Node *use = n->fast_out(i);
       if(use->is_Mem() && use->in(MemNode::Address) == n) {
-        memnode_worklist.push(use);
-      } else if (use->is_AddP() || use->is_Phi() || use->Opcode() == Op_CastPP || use->Opcode() == Op_CheckCastPP) {
-        alloc_worklist.push(use);
+        memnode_worklist.append_if_missing(use);
+      } else if (use->is_Initialize()) {
+        memnode_worklist.append_if_missing(use);
+      } else if (use->is_MergeMem()) {
+        mergemem_worklist.append_if_missing(use);
+      } else if (use->is_Call() && tinst != NULL) {
+        // Look for MergeMem nodes for calls which reference unique allocation
+        // (through CheckCastPP nodes) even for debug info.
+        Node* m = use->in(TypeFunc::Memory);
+        uint iid = tinst->instance_id();
+        while (m->is_Proj() && m->in(0)->is_Call() &&
+               m->in(0) != use && !m->in(0)->_idx != iid) {
+          m = m->in(0)->in(TypeFunc::Memory);
+        }
+        if (m->is_MergeMem()) {
+          mergemem_worklist.append_if_missing(m);
+        }
+      } else if (use->is_AddP() && use->outcnt() > 0) { // No dead nodes
+        Node* addp2 = find_second_addp(use, n);
+        if (addp2 != NULL) {
+          alloc_worklist.append_if_missing(addp2);
+        }
+        alloc_worklist.append_if_missing(use);
+      } else if (use->is_Phi() ||
+                 use->is_CheckCastPP() ||
+                 (use->is_ConstraintCast() && use->Opcode() == Op_CastPP)) {
+        alloc_worklist.append_if_missing(use);
       }
     }
 
   }
+  // New alias types were created in split_AddP().
   uint new_index_end = (uint) _compile->num_alias_types();
 
   //  Phase 2:  Process MemNode's from memnode_worklist. compute new address type and
@@ -668,32 +964,37 @@
   if (memnode_worklist.length() == 0)
     return;  // nothing to do
 
-
   while (memnode_worklist.length() != 0) {
     Node *n = memnode_worklist.pop();
+    if (visited.test_set(n->_idx))
+      continue;
     if (n->is_Phi()) {
       assert(n->as_Phi()->adr_type() != TypePtr::BOTTOM, "narrow memory slice required");
       // we don't need to do anything, but the users must be pushed if we haven't processed
       // this Phi before
-      if (visited.test_set(n->_idx))
+    } else if (n->is_Initialize()) {
+      // we don't need to do anything, but the users of the memory projection must be pushed
+      n = n->as_Initialize()->proj_out(TypeFunc::Memory);
+      if (n == NULL)
         continue;
     } else {
       assert(n->is_Mem(), "memory node required.");
       Node *addr = n->in(MemNode::Address);
+      assert(addr->is_AddP(), "AddP required");
       const Type *addr_t = igvn->type(addr);
       if (addr_t == Type::TOP)
         continue;
       assert (addr_t->isa_ptr() != NULL, "pointer type required.");
       int alias_idx = _compile->get_alias_index(addr_t->is_ptr());
-      Node *mem = find_mem(n->in(MemNode::Memory), alias_idx, igvn);
-      if (mem->is_Phi()) {
-        mem = split_memory_phi(mem->as_Phi(), alias_idx, orig_phis, igvn);
-      }
+      assert ((uint)alias_idx < new_index_end, "wrong alias index");
+      Node *mem = find_inst_mem(n->in(MemNode::Memory), alias_idx, orig_phis, igvn);
       if (_compile->failing()) {
         return;
       }
-      if (mem != n->in(MemNode::Memory))
+      if (mem != n->in(MemNode::Memory)) {
         set_map(n->_idx, mem);
+        _nodes->adr_at(n->_idx)->_node = n;
+      }
       if (n->is_Load()) {
         continue;  // don't push users
       } else if (n->is_LoadStore()) {
@@ -712,29 +1013,33 @@
     for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
       Node *use = n->fast_out(i);
       if (use->is_Phi()) {
-        memnode_worklist.push(use);
+        memnode_worklist.append_if_missing(use);
       } else if(use->is_Mem() && use->in(MemNode::Memory) == n) {
-        memnode_worklist.push(use);
+        memnode_worklist.append_if_missing(use);
+      } else if (use->is_Initialize()) {
+        memnode_worklist.append_if_missing(use);
       } else if (use->is_MergeMem()) {
-        mergemem_worklist.push(use);
+        mergemem_worklist.append_if_missing(use);
       }
     }
   }
 
-  //  Phase 3:  Process MergeMem nodes from mergemem_worklist.  Walk each memory slice
-  //            moving the first node encountered of each  instance type to the
-  //            the input corresponding to its alias index.
+  //  Phase 3:  Process MergeMem nodes from mergemem_worklist.
+  //            Walk each memory moving the first node encountered of each
+  //            instance type to the the input corresponding to its alias index.
   while (mergemem_worklist.length() != 0) {
     Node *n = mergemem_worklist.pop();
     assert(n->is_MergeMem(), "MergeMem node required.");
+    if (visited.test_set(n->_idx))
+      continue;
     MergeMemNode *nmm = n->as_MergeMem();
     // Note: we don't want to use MergeMemStream here because we only want to
-    //       scan inputs which exist at the start, not ones we add during processing
+    //  scan inputs which exist at the start, not ones we add during processing.
     uint nslices = nmm->req();
     igvn->hash_delete(nmm);
     for (uint i = Compile::AliasIdxRaw+1; i < nslices; i++) {
-      Node * mem = nmm->in(i);
-      Node * cur = NULL;
+      Node* mem = nmm->in(i);
+      Node* cur = NULL;
       if (mem == NULL || mem->is_top())
         continue;
       while (mem->is_Mem()) {
@@ -754,30 +1059,76 @@
         mem = mem->in(MemNode::Memory);
       }
       nmm->set_memory_at(i, (cur != NULL) ? cur : mem);
-      if (mem->is_Phi()) {
-        // We have encountered a Phi, we need to split the Phi for
-        // any  instance of the current type if we haven't encountered
-        //  a value of the instance along the chain.
-        for (uint ni = new_index_start; ni < new_index_end; ni++) {
-          if((uint)_compile->get_general_index(ni) == i) {
-            Node *m = (ni >= nmm->req()) ? nmm->empty_memory() : nmm->in(ni);
-            if (nmm->is_empty_memory(m)) {
-              m = split_memory_phi(mem->as_Phi(), ni, orig_phis, igvn);
-              if (_compile->failing()) {
-                return;
-              }
-              nmm->set_memory_at(ni, m);
+      // Find any instance of the current type if we haven't encountered
+      // a value of the instance along the chain.
+      for (uint ni = new_index_start; ni < new_index_end; ni++) {
+        if((uint)_compile->get_general_index(ni) == i) {
+          Node *m = (ni >= nmm->req()) ? nmm->empty_memory() : nmm->in(ni);
+          if (nmm->is_empty_memory(m)) {
+            Node* result = find_inst_mem(mem, ni, orig_phis, igvn);
+            if (_compile->failing()) {
+              return;
+            }
+            nmm->set_memory_at(ni, result);
+          }
+        }
+      }
+    }
+    // Find the rest of instances values
+    for (uint ni = new_index_start; ni < new_index_end; ni++) {
+      const TypeOopPtr *tinst = igvn->C->get_adr_type(ni)->isa_oopptr();
+      Node* result = step_through_mergemem(nmm, ni, tinst);
+      if (result == nmm->base_memory()) {
+        // Didn't find instance memory, search through general slice recursively.
+        result = nmm->memory_at(igvn->C->get_general_index(ni));
+        result = find_inst_mem(result, ni, orig_phis, igvn);
+        if (_compile->failing()) {
+          return;
+        }
+        nmm->set_memory_at(ni, result);
+      }
+    }
+    igvn->hash_insert(nmm);
+    record_for_optimizer(nmm);
+
+    // Propagate new memory slices to following MergeMem nodes.
+    for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
+      Node *use = n->fast_out(i);
+      if (use->is_Call()) {
+        CallNode* in = use->as_Call();
+        if (in->proj_out(TypeFunc::Memory) != NULL) {
+          Node* m = in->proj_out(TypeFunc::Memory);
+          for (DUIterator_Fast jmax, j = m->fast_outs(jmax); j < jmax; j++) {
+            Node* mm = m->fast_out(j);
+            if (mm->is_MergeMem()) {
+              mergemem_worklist.append_if_missing(mm);
+            }
+          }
+        }
+        if (use->is_Allocate()) {
+          use = use->as_Allocate()->initialization();
+          if (use == NULL) {
+            continue;
+          }
+        }
+      }
+      if (use->is_Initialize()) {
+        InitializeNode* in = use->as_Initialize();
+        if (in->proj_out(TypeFunc::Memory) != NULL) {
+          Node* m = in->proj_out(TypeFunc::Memory);
+          for (DUIterator_Fast jmax, j = m->fast_outs(jmax); j < jmax; j++) {
+            Node* mm = m->fast_out(j);
+            if (mm->is_MergeMem()) {
+              mergemem_worklist.append_if_missing(mm);
             }
           }
         }
       }
     }
-    igvn->hash_insert(nmm);
-    record_for_optimizer(nmm);
   }
 
-  //  Phase 4:  Update the inputs of non-instance memory Phis and the Memory input of memnodes
-  //
+  //  Phase 4:  Update the inputs of non-instance memory Phis and
+  //            the Memory input of memnodes
   // First update the inputs of any non-instance Phi's from
   // which we split out an instance Phi.  Note we don't have
   // to recursively process Phi's encounted on the input memory
@@ -789,7 +1140,10 @@
     igvn->hash_delete(phi);
     for (uint i = 1; i < phi->req(); i++) {
       Node *mem = phi->in(i);
-      Node *new_mem = find_mem(mem, alias_idx, igvn);
+      Node *new_mem = find_inst_mem(mem, alias_idx, orig_phis, igvn);
+      if (_compile->failing()) {
+        return;
+      }
       if (mem != new_mem) {
         phi->set_req(i, new_mem);
       }
@@ -803,7 +1157,7 @@
   for (int i = 0; i < _nodes->length(); i++) {
     Node *nmem = get_map(i);
     if (nmem != NULL) {
-      Node *n = _nodes->at(i)._node;
+      Node *n = _nodes->adr_at(i)->_node;
       if (n != NULL && n->is_Mem()) {
         igvn->hash_delete(n);
         n->set_req(MemNode::Memory, nmem);
@@ -815,59 +1169,110 @@
 }
 
 void ConnectionGraph::compute_escape() {
-  GrowableArray<int>  worklist;
-  GrowableArray<Node *>  alloc_worklist;
-  VectorSet visited(Thread::current()->resource_area());
-  PhaseGVN  *igvn = _compile->initial_gvn();
+
+  // 1. Populate Connection Graph with Ideal nodes.
+
+  Unique_Node_List worklist_init;
+  worklist_init.map(_compile->unique(), NULL);  // preallocate space
+
+  // Initialize worklist
+  if (_compile->root() != NULL) {
+    worklist_init.push(_compile->root());
+  }
+
+  GrowableArray<int> cg_worklist;
+  PhaseGVN* igvn = _compile->initial_gvn();
+  bool has_allocations = false;
+
+  // Push all useful nodes onto CG list and set their type.
+  for( uint next = 0; next < worklist_init.size(); ++next ) {
+    Node* n = worklist_init.at(next);
+    record_for_escape_analysis(n, igvn);
+    if (n->is_Call() &&
+        _nodes->adr_at(n->_idx)->node_type() == PointsToNode::JavaObject) {
+      has_allocations = true;
+    }
+    if(n->is_AddP())
+      cg_worklist.append(n->_idx);
+    for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
+      Node* m = n->fast_out(i);   // Get user
+      worklist_init.push(m);
+    }
+  }
 
-  // process Phi nodes from the deferred list, they may not have
-  while(_deferred.size() > 0) {
-    Node * n = _deferred.pop();
-    PhiNode * phi = n->as_Phi();
+  if (has_allocations) {
+    _has_allocations = true;
+  } else {
+    _has_allocations = false;
+    _collecting = false;
+    return; // Nothing to do.
+  }
+
+  // 2. First pass to create simple CG edges (doesn't require to walk CG).
+  for( uint next = 0; next < _delayed_worklist.size(); ++next ) {
+    Node* n = _delayed_worklist.at(next);
+    build_connection_graph(n, igvn);
+  }
 
-    process_phi_escape(phi, igvn);
+  // 3. Pass to create fields edges (Allocate -F-> AddP).
+  for( int next = 0; next < cg_worklist.length(); ++next ) {
+    int ni = cg_worklist.at(next);
+    build_connection_graph(_nodes->adr_at(ni)->_node, igvn);
+  }
+
+  cg_worklist.clear();
+  cg_worklist.append(_phantom_object);
+
+  // 4. Build Connection Graph which need
+  //    to walk the connection graph.
+  for (uint ni = 0; ni < (uint)_nodes->length(); ni++) {
+    PointsToNode* ptn = _nodes->adr_at(ni);
+    Node *n = ptn->_node;
+    if (n != NULL) { // Call, AddP, LoadP, StoreP
+      build_connection_graph(n, igvn);
+      if (ptn->node_type() != PointsToNode::UnknownType)
+        cg_worklist.append(n->_idx); // Collect CG nodes
+    }
   }
 
   VectorSet ptset(Thread::current()->resource_area());
+  GrowableArray<Node*>  alloc_worklist;
+  GrowableArray<int>  worklist;
 
   // remove deferred edges from the graph and collect
   // information we will need for type splitting
-  for (uint ni = 0; ni < (uint)_nodes->length(); ni++) {
-    PointsToNode * ptn = _nodes->adr_at(ni);
+  for( int next = 0; next < cg_worklist.length(); ++next ) {
+    int ni = cg_worklist.at(next);
+    PointsToNode* ptn = _nodes->adr_at(ni);
     PointsToNode::NodeType nt = ptn->node_type();
-
-    if (nt == PointsToNode::UnknownType) {
-      continue;  // not a node we are interested in
-    }
     Node *n = ptn->_node;
     if (nt == PointsToNode::LocalVar || nt == PointsToNode::Field) {
       remove_deferred(ni);
       if (n->is_AddP()) {
-        // if this AddP computes an address which may point to more that one
-        // object, nothing the address points to can be a unique type.
-        Node *base = n->in(AddPNode::Base);
+        // If this AddP computes an address which may point to more that one
+        // object, nothing the address points to can be scalar replaceable.
+        Node *base = get_addp_base(n);
         ptset.Clear();
         PointsTo(ptset, base, igvn);
         if (ptset.Size() > 1) {
           for( VectorSetI j(&ptset); j.test(); ++j ) {
-            PointsToNode *ptaddr = _nodes->adr_at(j.elem);
-            ptaddr->_unique_type = false;
+            uint pt = j.elem;
+            ptnode_adr(pt)->_scalar_replaceable = false;
           }
         }
       }
-    } else if (n->is_Call()) {
-        // initialize _escape_state of calls to GlobalEscape
-        n->as_Call()->_escape_state = PointsToNode::GlobalEscape;
-        // push call on alloc_worlist (alocations are calls)
-        // for processing by split_unique_types()
-        alloc_worklist.push(n);
+    } else if (nt == PointsToNode::JavaObject && n->is_Call()) {
+      // Push call on alloc_worlist (alocations are calls)
+      // for processing by split_unique_types().
+      alloc_worklist.append(n);
     }
   }
+
   // push all GlobalEscape nodes on the worklist
-  for (uint nj = 0; nj < (uint)_nodes->length(); nj++) {
-    if (_nodes->at(nj).escape_state() == PointsToNode::GlobalEscape) {
-      worklist.append(nj);
-    }
+  for( int next = 0; next < cg_worklist.length(); ++next ) {
+    int nk = cg_worklist.at(next);
+    if (_nodes->adr_at(nk)->escape_state() == PointsToNode::GlobalEscape)
+      worklist.append(nk);
   }
   // mark all node reachable from GlobalEscape nodes
   while(worklist.length() > 0) {
@@ -875,7 +1280,7 @@
     for (uint ei = 0; ei < n.edge_count(); ei++) {
       uint npi = n.edge_target(ei);
       PointsToNode *np = ptnode_adr(npi);
-      if (np->escape_state() != PointsToNode::GlobalEscape) {
+      if (np->escape_state() < PointsToNode::GlobalEscape) {
         np->set_escape_state(PointsToNode::GlobalEscape);
         worklist.append_if_missing(npi);
       }
@@ -883,133 +1288,191 @@
   }
 
   // push all ArgEscape nodes on the worklist
-  for (uint nk = 0; nk < (uint)_nodes->length(); nk++) {
-    if (_nodes->at(nk).escape_state() == PointsToNode::ArgEscape)
+  for( int next = 0; next < cg_worklist.length(); ++next ) {
+    int nk = cg_worklist.at(next);
+    if (_nodes->adr_at(nk)->escape_state() == PointsToNode::ArgEscape)
       worklist.push(nk);
   }
   // mark all node reachable from ArgEscape nodes
   while(worklist.length() > 0) {
     PointsToNode n = _nodes->at(worklist.pop());
-
     for (uint ei = 0; ei < n.edge_count(); ei++) {
       uint npi = n.edge_target(ei);
       PointsToNode *np = ptnode_adr(npi);
-      if (np->escape_state() != PointsToNode::ArgEscape) {
+      if (np->escape_state() < PointsToNode::ArgEscape) {
         np->set_escape_state(PointsToNode::ArgEscape);
         worklist.append_if_missing(npi);
       }
     }
   }
+
+  // push all NoEscape nodes on the worklist
+  for( int next = 0; next < cg_worklist.length(); ++next ) {
+    int nk = cg_worklist.at(next);
+    if (_nodes->adr_at(nk)->escape_state() == PointsToNode::NoEscape)
+      worklist.push(nk);
+  }
+  // mark all node reachable from NoEscape nodes
+  while(worklist.length() > 0) {
+    PointsToNode n = _nodes->at(worklist.pop());
+    for (uint ei = 0; ei < n.edge_count(); ei++) {
+      uint npi = n.edge_target(ei);
+      PointsToNode *np = ptnode_adr(npi);
+      if (np->escape_state() < PointsToNode::NoEscape) {
+        np->set_escape_state(PointsToNode::NoEscape);
+        worklist.append_if_missing(npi);
+      }
+    }
+  }
+
   _collecting = false;
 
-  // Now use the escape information to create unique types for
-  // unescaped objects
-  split_unique_types(alloc_worklist);
-  if (_compile->failing())  return;
-
-  // Clean up after split unique types.
-  ResourceMark rm;
-  PhaseRemoveUseless pru(_compile->initial_gvn(), _compile->for_igvn());
-}
+  has_allocations = false; // Are there scalar replaceable allocations?
 
-Node * ConnectionGraph::skip_casts(Node *n) {
-  while(n->Opcode() == Op_CastPP || n->Opcode() == Op_CheckCastPP) {
-    n = n->in(1);
+  for( int next = 0; next < alloc_worklist.length(); ++next ) {
+    Node* n = alloc_worklist.at(next);
+    uint ni = n->_idx;
+    PointsToNode* ptn = _nodes->adr_at(ni);
+    PointsToNode::EscapeState es = ptn->escape_state();
+    if (ptn->escape_state() == PointsToNode::NoEscape &&
+        ptn->_scalar_replaceable) {
+      has_allocations = true;
+      break;
+    }
   }
-  return n;
-}
-
-void ConnectionGraph::process_phi_escape(PhiNode *phi, PhaseTransform *phase) {
-
-  if (phi->type()->isa_oopptr() == NULL)
-    return;  // nothing to do if not an oop
+  if (!has_allocations) {
+    return; // Nothing to do.
+  }
 
-  PointsToNode *ptadr = ptnode_adr(phi->_idx);
-  int incount = phi->req();
-  int non_null_inputs = 0;
+  if(_compile->AliasLevel() >= 3 && EliminateAllocations) {
+    // Now use the escape information to create unique types for
+    // unescaped objects
+    split_unique_types(alloc_worklist);
+    if (_compile->failing())  return;
 
-  for (int i = 1; i < incount ; i++) {
-    if (phi->in(i) != NULL)
-      non_null_inputs++;
-  }
-  if (non_null_inputs == ptadr->_inputs_processed)
-    return;  // no new inputs since the last time this node was processed,
-             // the current information is valid
+    // Clean up after split unique types.
+    ResourceMark rm;
+    PhaseRemoveUseless pru(_compile->initial_gvn(), _compile->for_igvn());
 
-  ptadr->_inputs_processed = non_null_inputs;  // prevent recursive processing of this node
-  for (int j = 1; j < incount ; j++) {
-    Node * n = phi->in(j);
-    if (n == NULL)
-      continue;  // ignore NULL
-    n =  skip_casts(n);
-    if (n->is_top() || n == phi)
-      continue;  // ignore top or inputs which go back this node
-    int nopc = n->Opcode();
-    PointsToNode  npt = _nodes->at(n->_idx);
-    if (_nodes->at(n->_idx).node_type() == PointsToNode::JavaObject) {
-      add_pointsto_edge(phi->_idx, n->_idx);
-    } else {
-      add_deferred_edge(phi->_idx, n->_idx);
+#ifdef ASSERT
+  } else if (PrintEscapeAnalysis || PrintEliminateAllocations) {
+    tty->print("=== No allocations eliminated for ");
+    C()->method()->print_short_name();
+    if(!EliminateAllocations) {
+      tty->print(" since EliminateAllocations is off ===");
+    } else if(_compile->AliasLevel() < 3) {
+      tty->print(" since AliasLevel < 3 ===");
     }
+    tty->cr();
+#endif
   }
 }
 
 void ConnectionGraph::process_call_arguments(CallNode *call, PhaseTransform *phase) {
 
-    _processed.set(call->_idx);
     switch (call->Opcode()) {
-
-    // arguments to allocation and locking don't escape
+#ifdef ASSERT
     case Op_Allocate:
     case Op_AllocateArray:
     case Op_Lock:
     case Op_Unlock:
+      assert(false, "should be done already");
       break;
+#endif
+    case Op_CallLeafNoFP:
+    {
+      // Stub calls, objects do not escape but they are not scale replaceable.
+      // Adjust escape state for outgoing arguments.
+      const TypeTuple * d = call->tf()->domain();
+      VectorSet ptset(Thread::current()->resource_area());
+      for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
+        const Type* at = d->field_at(i);
+        Node *arg = call->in(i)->uncast();
+        const Type *aat = phase->type(arg);
+        if (!arg->is_top() && at->isa_ptr() && aat->isa_ptr()) {
+          assert(aat == Type::TOP || aat == TypePtr::NULL_PTR ||
+                 aat->isa_ptr() != NULL, "expecting an Ptr");
+          set_escape_state(arg->_idx, PointsToNode::ArgEscape);
+          if (arg->is_AddP()) {
+            //
+            // The inline_native_clone() case when the arraycopy stub is called
+            // after the allocation before Initialize and CheckCastPP nodes.
+            //
+            // Set AddP's base (Allocate) as not scalar replaceable since
+            // pointer to the base (with offset) is passed as argument.
+            //
+            arg = get_addp_base(arg);
+          }
+          ptset.Clear();
+          PointsTo(ptset, arg, phase);
+          for( VectorSetI j(&ptset); j.test(); ++j ) {
+            uint pt = j.elem;
+            set_escape_state(pt, PointsToNode::ArgEscape);
+          }
+        }
+      }
+      break;
+    }
 
     case Op_CallStaticJava:
     // For a static call, we know exactly what method is being called.
     // Use bytecode estimator to record the call's escape affects
     {
       ciMethod *meth = call->as_CallJava()->method();
-      if (meth != NULL) {
+      BCEscapeAnalyzer *call_analyzer = (meth !=NULL) ? meth->get_bcea() : NULL;
+      // fall-through if not a Java method or no analyzer information
+      if (call_analyzer != NULL) {
         const TypeTuple * d = call->tf()->domain();
-        BCEscapeAnalyzer call_analyzer(meth);
         VectorSet ptset(Thread::current()->resource_area());
+        bool copy_dependencies = false;
         for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
           const Type* at = d->field_at(i);
           int k = i - TypeFunc::Parms;
 
           if (at->isa_oopptr() != NULL) {
-            Node *arg = skip_casts(call->in(i));
-
-            if (!call_analyzer.is_arg_stack(k)) {
-              // The argument global escapes, mark everything it could point to
-              ptset.Clear();
-              PointsTo(ptset, arg, phase);
-              for( VectorSetI j(&ptset); j.test(); ++j ) {
-                uint pt = j.elem;
+            Node *arg = call->in(i)->uncast();
 
-                set_escape_state(pt, PointsToNode::GlobalEscape);
+            bool global_escapes = false;
+            bool fields_escapes = false;
+            if (!call_analyzer->is_arg_stack(k)) {
+              // The argument global escapes, mark everything it could point to
+              set_escape_state(arg->_idx, PointsToNode::GlobalEscape);
+              global_escapes = true;
+            } else {
+              if (!call_analyzer->is_arg_local(k)) {
+                // The argument itself doesn't escape, but any fields might
+                fields_escapes = true;
               }
-            } else if (!call_analyzer.is_arg_local(k)) {
-              // The argument itself doesn't escape, but any fields might
-              ptset.Clear();
-              PointsTo(ptset, arg, phase);
-              for( VectorSetI j(&ptset); j.test(); ++j ) {
-                uint pt = j.elem;
-                add_edge_from_fields(pt, _phantom_object, Type::OffsetBot);
+              set_escape_state(arg->_idx, PointsToNode::ArgEscape);
+              copy_dependencies = true;
+            }
+
+            ptset.Clear();
+            PointsTo(ptset, arg, phase);
+            for( VectorSetI j(&ptset); j.test(); ++j ) {
+              uint pt = j.elem;
+              if (global_escapes) {
+                //The argument global escapes, mark everything it could point to
+                set_escape_state(pt, PointsToNode::GlobalEscape);
+              } else {
+                if (fields_escapes) {
+                  // The argument itself doesn't escape, but any fields might
+                  add_edge_from_fields(pt, _phantom_object, Type::OffsetBot);
+                }
+                set_escape_state(pt, PointsToNode::ArgEscape);
               }
             }
           }
         }
-        call_analyzer.copy_dependencies(C()->dependencies());
+        if (copy_dependencies)
+          call_analyzer->copy_dependencies(C()->dependencies());
         break;
       }
-      // fall-through if not a Java method
     }
 
     default:
-    // Some other type of call, assume the worst case: all arguments
+    // Fall-through here if not a Java method or no analyzer information
+    // or some other type of call, assume the worst case: all arguments
     // globally escape.
     {
       // adjust escape state for  outgoing arguments
@@ -1017,15 +1480,15 @@
       VectorSet ptset(Thread::current()->resource_area());
       for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
         const Type* at = d->field_at(i);
-
         if (at->isa_oopptr() != NULL) {
-          Node *arg = skip_casts(call->in(i));
+          Node *arg = call->in(i)->uncast();
+          set_escape_state(arg->_idx, PointsToNode::GlobalEscape);
           ptset.Clear();
           PointsTo(ptset, arg, phase);
           for( VectorSetI j(&ptset); j.test(); ++j ) {
             uint pt = j.elem;
-
             set_escape_state(pt, PointsToNode::GlobalEscape);
+            PointsToNode *ptadr = ptnode_adr(pt);
           }
         }
       }
@@ -1033,15 +1496,9 @@
   }
 }
 void ConnectionGraph::process_call_result(ProjNode *resproj, PhaseTransform *phase) {
-  CallNode *call = resproj->in(0)->as_Call();
-
   PointsToNode *ptadr = ptnode_adr(resproj->_idx);
 
-  ptadr->_node = resproj;
-  ptadr->set_node_type(PointsToNode::LocalVar);
-  set_escape_state(resproj->_idx, PointsToNode::UnknownEscape);
-  _processed.set(resproj->_idx);
-
+  CallNode *call = resproj->in(0)->as_Call();
   switch (call->Opcode()) {
     case Op_Allocate:
     {
@@ -1057,36 +1514,40 @@
       ciInstanceKlass* ciik = cik->as_instance_klass();
 
       PointsToNode *ptadr = ptnode_adr(call->_idx);
-      ptadr->set_node_type(PointsToNode::JavaObject);
+      PointsToNode::EscapeState es;
+      uint edge_to;
       if (cik->is_subclass_of(_compile->env()->Thread_klass()) || ciik->has_finalizer()) {
-        set_escape_state(call->_idx, PointsToNode::GlobalEscape);
-        add_pointsto_edge(resproj->_idx, _phantom_object);
+        es = PointsToNode::GlobalEscape;
+        edge_to = _phantom_object; // Could not be worse
       } else {
-        set_escape_state(call->_idx, PointsToNode::NoEscape);
-        add_pointsto_edge(resproj->_idx, call->_idx);
+        es = PointsToNode::NoEscape;
+        edge_to = call->_idx;
       }
-      _processed.set(call->_idx);
+      set_escape_state(call->_idx, es);
+      add_pointsto_edge(resproj->_idx, edge_to);
+      _processed.set(resproj->_idx);
       break;
     }
 
     case Op_AllocateArray:
     {
       PointsToNode *ptadr = ptnode_adr(call->_idx);
-      ptadr->set_node_type(PointsToNode::JavaObject);
+      int length = call->in(AllocateNode::ALength)->find_int_con(-1);
+      if (length < 0 || length > EliminateAllocationArraySizeLimit) {
+        // Not scalar replaceable if the length is not constant or too big.
+        ptadr->_scalar_replaceable = false;
+      }
       set_escape_state(call->_idx, PointsToNode::NoEscape);
-      _processed.set(call->_idx);
       add_pointsto_edge(resproj->_idx, call->_idx);
+      _processed.set(resproj->_idx);
       break;
     }
 
-    case Op_Lock:
-    case Op_Unlock:
-      break;
-
     case Op_CallStaticJava:
     // For a static call, we know exactly what method is being called.
     // Use bytecode estimator to record whether the call's return value escapes
     {
+      bool done = true;
       const TypeTuple *r = call->tf()->range();
       const Type* ret_type = NULL;
 
@@ -1095,32 +1556,45 @@
 
       // Note:  we use isa_ptr() instead of isa_oopptr()  here because the
       //        _multianewarray functions return a TypeRawPtr.
-      if (ret_type == NULL || ret_type->isa_ptr() == NULL)
+      if (ret_type == NULL || ret_type->isa_ptr() == NULL) {
+        _processed.set(resproj->_idx);
         break;  // doesn't return a pointer type
-
+      }
       ciMethod *meth = call->as_CallJava()->method();
+      const TypeTuple * d = call->tf()->domain();
       if (meth == NULL) {
         // not a Java method, assume global escape
         set_escape_state(call->_idx, PointsToNode::GlobalEscape);
         if (resproj != NULL)
           add_pointsto_edge(resproj->_idx, _phantom_object);
       } else {
-        BCEscapeAnalyzer call_analyzer(meth);
+        BCEscapeAnalyzer *call_analyzer = meth->get_bcea();
         VectorSet ptset(Thread::current()->resource_area());
+        bool copy_dependencies = false;
 
-        if (call_analyzer.is_return_local() && resproj != NULL) {
+        if (call_analyzer->is_return_allocated()) {
+          // Returns a newly allocated unescaped object, simply
+          // update dependency information.
+          // Mark it as NoEscape so that objects referenced by
+          // it's fields will be marked as NoEscape at least.
+          set_escape_state(call->_idx, PointsToNode::NoEscape);
+          if (resproj != NULL)
+            add_pointsto_edge(resproj->_idx, call->_idx);
+          copy_dependencies = true;
+        } else if (call_analyzer->is_return_local() && resproj != NULL) {
           // determine whether any arguments are returned
-          const TypeTuple * d = call->tf()->domain();
           set_escape_state(call->_idx, PointsToNode::NoEscape);
           for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
             const Type* at = d->field_at(i);
 
             if (at->isa_oopptr() != NULL) {
-              Node *arg = skip_casts(call->in(i));
+              Node *arg = call->in(i)->uncast();
 
-              if (call_analyzer.is_arg_returned(i - TypeFunc::Parms)) {
+              if (call_analyzer->is_arg_returned(i - TypeFunc::Parms)) {
                 PointsToNode *arg_esp = _nodes->adr_at(arg->_idx);
-                if (arg_esp->node_type() == PointsToNode::JavaObject)
+                if (arg_esp->node_type() == PointsToNode::UnknownType)
+                  done = false;
+                else if (arg_esp->node_type() == PointsToNode::JavaObject)
                   add_pointsto_edge(resproj->_idx, arg->_idx);
                 else
                   add_deferred_edge(resproj->_idx, arg->_idx);
@@ -1128,13 +1602,25 @@
               }
             }
           }
+          copy_dependencies = true;
         } else {
           set_escape_state(call->_idx, PointsToNode::GlobalEscape);
           if (resproj != NULL)
             add_pointsto_edge(resproj->_idx, _phantom_object);
+          for (uint i = TypeFunc::Parms; i < d->cnt(); i++) {
+            const Type* at = d->field_at(i);
+            if (at->isa_oopptr() != NULL) {
+              Node *arg = call->in(i)->uncast();
+              PointsToNode *arg_esp = _nodes->adr_at(arg->_idx);
+              arg_esp->_hidden_alias = true;
+            }
+          }
         }
-        call_analyzer.copy_dependencies(C()->dependencies());
+        if (copy_dependencies)
+          call_analyzer->copy_dependencies(C()->dependencies());
       }
+      if (done)
+        _processed.set(resproj->_idx);
       break;
     }
 
@@ -1143,7 +1629,6 @@
     // returned value, if any, globally escapes.
     {
       const TypeTuple *r = call->tf()->range();
-
       if (r->cnt() > TypeFunc::Parms) {
         const Type* ret_type = r->field_at(TypeFunc::Parms);
 
@@ -1151,142 +1636,385 @@
         //        _multianewarray functions return a TypeRawPtr.
         if (ret_type->isa_ptr() != NULL) {
           PointsToNode *ptadr = ptnode_adr(call->_idx);
-          ptadr->set_node_type(PointsToNode::JavaObject);
           set_escape_state(call->_idx, PointsToNode::GlobalEscape);
           if (resproj != NULL)
             add_pointsto_edge(resproj->_idx, _phantom_object);
         }
       }
-    }
-  }
-}
-
-void ConnectionGraph::record_for_escape_analysis(Node *n) {
-  if (_collecting) {
-    if (n->is_Phi()) {
-      PhiNode *phi = n->as_Phi();
-      const Type *pt = phi->type();
-      if ((pt->isa_oopptr() != NULL) || pt == TypePtr::NULL_PTR) {
-        PointsToNode *ptn = ptnode_adr(phi->_idx);
-        ptn->set_node_type(PointsToNode::LocalVar);
-        ptn->_node = n;
-        _deferred.push(n);
-      }
+      _processed.set(resproj->_idx);
     }
   }
 }
 
-void ConnectionGraph::record_escape_work(Node *n, PhaseTransform *phase) {
+// Populate Connection Graph with Ideal nodes and create simple
+// connection graph edges (do not need to check the node_type of inputs
+// or to call PointsTo() to walk the connection graph).
+void ConnectionGraph::record_for_escape_analysis(Node *n, PhaseTransform *phase) {
+  if (_processed.test(n->_idx))
+    return; // No need to redefine node's state.
+
+  if (n->is_Call()) {
+    // Arguments to allocation and locking don't escape.
+    if (n->is_Allocate()) {
+      add_node(n, PointsToNode::JavaObject, PointsToNode::UnknownEscape, true);
+      record_for_optimizer(n);
+    } else if (n->is_Lock() || n->is_Unlock()) {
+      // Put Lock and Unlock nodes on IGVN worklist to process them during
+      // the first IGVN optimization when escape information is still available.
+      record_for_optimizer(n);
+      _processed.set(n->_idx);
+    } else {
+      // Have to process call's arguments first.
+      PointsToNode::NodeType nt = PointsToNode::UnknownType;
+
+      // Check if a call returns an object.
+      const TypeTuple *r = n->as_Call()->tf()->range();
+      if (r->cnt() > TypeFunc::Parms &&
+          n->as_Call()->proj_out(TypeFunc::Parms) != NULL) {
+        // Note:  use isa_ptr() instead of isa_oopptr() here because
+        //        the _multianewarray functions return a TypeRawPtr.
+        if (r->field_at(TypeFunc::Parms)->isa_ptr() != NULL) {
+          nt = PointsToNode::JavaObject;
+        }
+      }
+      add_node(n, nt, PointsToNode::UnknownEscape, false);
+    }
+    return;
+  }
+
+  // Using isa_ptr() instead of isa_oopptr() for LoadP and Phi because
+  // ThreadLocal has RawPrt type.
+  switch (n->Opcode()) {
+    case Op_AddP:
+    {
+      add_node(n, PointsToNode::Field, PointsToNode::UnknownEscape, false);
+      break;
+    }
+    case Op_CastX2P:
+    { // "Unsafe" memory access.
+      add_node(n, PointsToNode::JavaObject, PointsToNode::GlobalEscape, true);
+      break;
+    }
+    case Op_CastPP:
+    case Op_CheckCastPP:
+    {
+      add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false);
+      int ti = n->in(1)->_idx;
+      PointsToNode::NodeType nt = _nodes->adr_at(ti)->node_type();
+      if (nt == PointsToNode::UnknownType) {
+        _delayed_worklist.push(n); // Process it later.
+        break;
+      } else if (nt == PointsToNode::JavaObject) {
+        add_pointsto_edge(n->_idx, ti);
+      } else {
+        add_deferred_edge(n->_idx, ti);
+      }
+      _processed.set(n->_idx);
+      break;
+    }
+    case Op_ConP:
+    {
+      // assume all pointer constants globally escape except for null
+      PointsToNode::EscapeState es;
+      if (phase->type(n) == TypePtr::NULL_PTR)
+        es = PointsToNode::NoEscape;
+      else
+        es = PointsToNode::GlobalEscape;
 
-  int opc = n->Opcode();
+      add_node(n, PointsToNode::JavaObject, es, true);
+      break;
+    }
+    case Op_CreateEx:
+    {
+      // assume that all exception objects globally escape
+      add_node(n, PointsToNode::JavaObject, PointsToNode::GlobalEscape, true);
+      break;
+    }
+    case Op_LoadKlass:
+    {
+      add_node(n, PointsToNode::JavaObject, PointsToNode::GlobalEscape, true);
+      break;
+    }
+    case Op_LoadP:
+    {
+      const Type *t = phase->type(n);
+      if (t->isa_ptr() == NULL) {
+        _processed.set(n->_idx);
+        return;
+      }
+      add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false);
+      break;
+    }
+    case Op_Parm:
+    {
+      _processed.set(n->_idx); // No need to redefine it state.
+      uint con = n->as_Proj()->_con;
+      if (con < TypeFunc::Parms)
+        return;
+      const Type *t = n->in(0)->as_Start()->_domain->field_at(con);
+      if (t->isa_ptr() == NULL)
+        return;
+      // We have to assume all input parameters globally escape
+      // (Note: passing 'false' since _processed is already set).
+      add_node(n, PointsToNode::JavaObject, PointsToNode::GlobalEscape, false);
+      break;
+    }
+    case Op_Phi:
+    {
+      if (n->as_Phi()->type()->isa_ptr() == NULL) {
+        // nothing to do if not an oop
+        _processed.set(n->_idx);
+        return;
+      }
+      add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false);
+      uint i;
+      for (i = 1; i < n->req() ; i++) {
+        Node* in = n->in(i);
+        if (in == NULL)
+          continue;  // ignore NULL
+        in = in->uncast();
+        if (in->is_top() || in == n)
+          continue;  // ignore top or inputs which go back this node
+        int ti = in->_idx;
+        PointsToNode::NodeType nt = _nodes->adr_at(ti)->node_type();
+        if (nt == PointsToNode::UnknownType) {
+          break;
+        } else if (nt == PointsToNode::JavaObject) {
+          add_pointsto_edge(n->_idx, ti);
+        } else {
+          add_deferred_edge(n->_idx, ti);
+        }
+      }
+      if (i >= n->req())
+        _processed.set(n->_idx);
+      else
+        _delayed_worklist.push(n);
+      break;
+    }
+    case Op_Proj:
+    {
+      // we are only interested in the result projection from a call
+      if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() ) {
+        add_node(n, PointsToNode::LocalVar, PointsToNode::UnknownEscape, false);
+        process_call_result(n->as_Proj(), phase);
+        if (!_processed.test(n->_idx)) {
+          // The call's result may need to be processed later if the call
+          // returns it's argument and the argument is not processed yet.
+          _delayed_worklist.push(n);
+        }
+      } else {
+        _processed.set(n->_idx);
+      }
+      break;
+    }
+    case Op_Return:
+    {
+      if( n->req() > TypeFunc::Parms &&
+          phase->type(n->in(TypeFunc::Parms))->isa_oopptr() ) {
+        // Treat Return value as LocalVar with GlobalEscape escape state.
+        add_node(n, PointsToNode::LocalVar, PointsToNode::GlobalEscape, false);
+        int ti = n->in(TypeFunc::Parms)->_idx;
+        PointsToNode::NodeType nt = _nodes->adr_at(ti)->node_type();
+        if (nt == PointsToNode::UnknownType) {
+          _delayed_worklist.push(n); // Process it later.
+          break;
+        } else if (nt == PointsToNode::JavaObject) {
+          add_pointsto_edge(n->_idx, ti);
+        } else {
+          add_deferred_edge(n->_idx, ti);
+        }
+      }
+      _processed.set(n->_idx);
+      break;
+    }
+    case Op_StoreP:
+    {
+      const Type *adr_type = phase->type(n->in(MemNode::Address));
+      if (adr_type->isa_oopptr()) {
+        add_node(n, PointsToNode::UnknownType, PointsToNode::UnknownEscape, false);
+      } else {
+        Node* adr = n->in(MemNode::Address);
+        if (adr->is_AddP() && phase->type(adr) == TypeRawPtr::NOTNULL &&
+            adr->in(AddPNode::Address)->is_Proj() &&
+            adr->in(AddPNode::Address)->in(0)->is_Allocate()) {
+          add_node(n, PointsToNode::UnknownType, PointsToNode::UnknownEscape, false);
+          // We are computing a raw address for a store captured
+          // by an Initialize compute an appropriate address type.
+          int offs = (int)phase->find_intptr_t_con(adr->in(AddPNode::Offset), Type::OffsetBot);
+          assert(offs != Type::OffsetBot, "offset must be a constant");
+        } else {
+          _processed.set(n->_idx);
+          return;
+        }
+      }
+      break;
+    }
+    case Op_StorePConditional:
+    case Op_CompareAndSwapP:
+    {
+      const Type *adr_type = phase->type(n->in(MemNode::Address));
+      if (adr_type->isa_oopptr()) {
+        add_node(n, PointsToNode::UnknownType, PointsToNode::UnknownEscape, false);
+      } else {
+        _processed.set(n->_idx);
+        return;
+      }
+      break;
+    }
+    case Op_ThreadLocal:
+    {
+      add_node(n, PointsToNode::JavaObject, PointsToNode::ArgEscape, true);
+      break;
+    }
+    default:
+      ;
+      // nothing to do
+  }
+  return;
+}
+
+void ConnectionGraph::build_connection_graph(Node *n, PhaseTransform *phase) {
+  // Don't set processed bit for AddP, LoadP, StoreP since
+  // they may need more then one pass to process.
+  if (_processed.test(n->_idx))
+    return; // No need to redefine node's state.
+
   PointsToNode *ptadr = ptnode_adr(n->_idx);
 
-  if (_processed.test(n->_idx))
-    return;
-
-  ptadr->_node = n;
   if (n->is_Call()) {
     CallNode *call = n->as_Call();
     process_call_arguments(call, phase);
+    _processed.set(n->_idx);
     return;
   }
 
-  switch (opc) {
+  switch (n->Opcode()) {
     case Op_AddP:
     {
-      Node *base = skip_casts(n->in(AddPNode::Base));
-      ptadr->set_node_type(PointsToNode::Field);
-
-      // create a field edge to this node from everything adr could point to
+      Node *base = get_addp_base(n);
+      // Create a field edge to this node from everything base could point to.
       VectorSet ptset(Thread::current()->resource_area());
       PointsTo(ptset, base, phase);
       for( VectorSetI i(&ptset); i.test(); ++i ) {
         uint pt = i.elem;
-        add_field_edge(pt, n->_idx, type_to_offset(phase->type(n)));
+        add_field_edge(pt, n->_idx, address_offset(n, phase));
+      }
+      break;
+    }
+    case Op_CastX2P:
+    {
+      assert(false, "Op_CastX2P");
+      break;
+    }
+    case Op_CastPP:
+    case Op_CheckCastPP:
+    {
+      int ti = n->in(1)->_idx;
+      if (_nodes->adr_at(ti)->node_type() == PointsToNode::JavaObject) {
+        add_pointsto_edge(n->_idx, ti);
+      } else {
+        add_deferred_edge(n->_idx, ti);
+      }
+      _processed.set(n->_idx);
+      break;
+    }
+    case Op_ConP:
+    {
+      assert(false, "Op_ConP");
+      break;
+    }
+    case Op_CreateEx:
+    {
+      assert(false, "Op_CreateEx");
+      break;
+    }
+    case Op_LoadKlass:
+    {
+      assert(false, "Op_LoadKlass");
+      break;
+    }
+    case Op_LoadP:
+    {
+      const Type *t = phase->type(n);
+#ifdef ASSERT
+      if (t->isa_ptr() == NULL)
+        assert(false, "Op_LoadP");
+#endif
+
+      Node* adr = n->in(MemNode::Address)->uncast();
+      const Type *adr_type = phase->type(adr);
+      Node* adr_base;
+      if (adr->is_AddP()) {
+        adr_base = get_addp_base(adr);
+      } else {
+        adr_base = adr;
+      }
+
+      // For everything "adr_base" could point to, create a deferred edge from
+      // this node to each field with the same offset.
+      VectorSet ptset(Thread::current()->resource_area());
+      PointsTo(ptset, adr_base, phase);
+      int offset = address_offset(adr, phase);
+      for( VectorSetI i(&ptset); i.test(); ++i ) {
+        uint pt = i.elem;
+        add_deferred_edge_to_fields(n->_idx, pt, offset);
       }
       break;
     }
     case Op_Parm:
     {
-      ProjNode *nproj = n->as_Proj();
-      uint con = nproj->_con;
-      if (con < TypeFunc::Parms)
-        return;
-      const Type *t = nproj->in(0)->as_Start()->_domain->field_at(con);
-      if (t->isa_ptr() == NULL)
-        return;
-      ptadr->set_node_type(PointsToNode::JavaObject);
-      if (t->isa_oopptr() != NULL) {
-        set_escape_state(n->_idx, PointsToNode::ArgEscape);
-      } else {
-        // this must be the incoming state of an OSR compile, we have to assume anything
-        // passed in globally escapes
-        assert(_compile->is_osr_compilation(), "bad argument type for non-osr compilation");
-        set_escape_state(n->_idx, PointsToNode::GlobalEscape);
-      }
-      _processed.set(n->_idx);
+      assert(false, "Op_Parm");
       break;
     }
     case Op_Phi:
     {
-      PhiNode *phi = n->as_Phi();
-      if (phi->type()->isa_oopptr() == NULL)
-        return;  // nothing to do if not an oop
-      ptadr->set_node_type(PointsToNode::LocalVar);
-      process_phi_escape(phi, phase);
-      break;
-    }
-    case Op_CreateEx:
-    {
-      // assume that all exception objects globally escape
-      ptadr->set_node_type(PointsToNode::JavaObject);
-      set_escape_state(n->_idx, PointsToNode::GlobalEscape);
-      _processed.set(n->_idx);
-      break;
-    }
-    case Op_ConP:
-    {
-      const Type *t = phase->type(n);
-      ptadr->set_node_type(PointsToNode::JavaObject);
-      // assume all pointer constants globally escape except for null
-      if (t == TypePtr::NULL_PTR)
-        set_escape_state(n->_idx, PointsToNode::NoEscape);
-      else
-        set_escape_state(n->_idx, PointsToNode::GlobalEscape);
+#ifdef ASSERT
+      if (n->as_Phi()->type()->isa_ptr() == NULL)
+        assert(false, "Op_Phi");
+#endif
+      for (uint i = 1; i < n->req() ; i++) {
+        Node* in = n->in(i);
+        if (in == NULL)
+          continue;  // ignore NULL
+        in = in->uncast();
+        if (in->is_top() || in == n)
+          continue;  // ignore top or inputs which go back this node
+        int ti = in->_idx;
+        if (_nodes->adr_at(in->_idx)->node_type() == PointsToNode::JavaObject) {
+          add_pointsto_edge(n->_idx, ti);
+        } else {
+          add_deferred_edge(n->_idx, ti);
+        }
+      }
       _processed.set(n->_idx);
       break;
     }
-    case Op_LoadKlass:
+    case Op_Proj:
     {
-      ptadr->set_node_type(PointsToNode::JavaObject);
-      set_escape_state(n->_idx, PointsToNode::GlobalEscape);
-      _processed.set(n->_idx);
+      // we are only interested in the result projection from a call
+      if (n->as_Proj()->_con == TypeFunc::Parms && n->in(0)->is_Call() ) {
+        process_call_result(n->as_Proj(), phase);
+        assert(_processed.test(n->_idx), "all call results should be processed");
+      } else {
+        assert(false, "Op_Proj");
+      }
       break;
     }
-    case Op_LoadP:
+    case Op_Return:
     {
-      const Type *t = phase->type(n);
-      if (!t->isa_oopptr())
-        return;
-      ptadr->set_node_type(PointsToNode::LocalVar);
-      set_escape_state(n->_idx, PointsToNode::UnknownEscape);
-
-      Node *adr = skip_casts(n->in(MemNode::Address));
-      const Type *adr_type = phase->type(adr);
-      Node *adr_base = skip_casts((adr->Opcode() == Op_AddP) ? adr->in(AddPNode::Base) : adr);
-
-      // For everything "adr" could point to, create a deferred edge from
-      // this node to each field with the same offset as "adr_type"
-      VectorSet ptset(Thread::current()->resource_area());
-      PointsTo(ptset, adr_base, phase);
-      // If ptset is empty, then this value must have been set outside
-      // this method, so we add the phantom node
-      if (ptset.Size() == 0)
-        ptset.set(_phantom_object);
-      for( VectorSetI i(&ptset); i.test(); ++i ) {
-        uint pt = i.elem;
-        add_deferred_edge_to_fields(n->_idx, pt, type_to_offset(adr_type));
+#ifdef ASSERT
+      if( n->req() <= TypeFunc::Parms ||
+          !phase->type(n->in(TypeFunc::Parms))->isa_oopptr() ) {
+        assert(false, "Op_Return");
       }
+#endif
+      int ti = n->in(TypeFunc::Parms)->_idx;
+      if (_nodes->adr_at(ti)->node_type() == PointsToNode::JavaObject) {
+        add_pointsto_edge(n->_idx, ti);
+      } else {
+        add_deferred_edge(n->_idx, ti);
+      }
+      _processed.set(n->_idx);
       break;
     }
     case Op_StoreP:
@@ -1294,45 +2022,28 @@
     case Op_CompareAndSwapP:
     {
       Node *adr = n->in(MemNode::Address);
-      Node *val = skip_casts(n->in(MemNode::ValueIn));
       const Type *adr_type = phase->type(adr);
+#ifdef ASSERT
       if (!adr_type->isa_oopptr())
-        return;
+        assert(phase->type(adr) == TypeRawPtr::NOTNULL, "Op_StoreP");
+#endif
 
-      assert(adr->Opcode() == Op_AddP, "expecting an AddP");
-      Node *adr_base = adr->in(AddPNode::Base);
-
-      // For everything "adr_base" could point to, create a deferred edge to "val" from each field
-      // with the same offset as "adr_type"
+      assert(adr->is_AddP(), "expecting an AddP");
+      Node *adr_base = get_addp_base(adr);
+      Node *val = n->in(MemNode::ValueIn)->uncast();
+      // For everything "adr_base" could point to, create a deferred edge
+      // to "val" from each field with the same offset.
       VectorSet ptset(Thread::current()->resource_area());
       PointsTo(ptset, adr_base, phase);
       for( VectorSetI i(&ptset); i.test(); ++i ) {
         uint pt = i.elem;
-        add_edge_from_fields(pt, val->_idx, type_to_offset(adr_type));
+        add_edge_from_fields(pt, val->_idx, address_offset(adr, phase));
       }
       break;
     }
-    case Op_Proj:
+    case Op_ThreadLocal:
     {
-      ProjNode *nproj = n->as_Proj();
-      Node *n0 = nproj->in(0);
-      // we are only interested in the result projection from a call
-      if (nproj->_con == TypeFunc::Parms && n0->is_Call() ) {
-        process_call_result(nproj, phase);
-      }
-
-      break;
-    }
-    case Op_CastPP:
-    case Op_CheckCastPP:
-    {
-      ptadr->set_node_type(PointsToNode::LocalVar);
-      int ti = n->in(1)->_idx;
-      if (_nodes->at(ti).node_type() == PointsToNode::JavaObject) {
-        add_pointsto_edge(n->_idx, ti);
-      } else {
-        add_deferred_edge(n->_idx, ti);
-      }
+      assert(false, "Op_ThreadLocal");
       break;
     }
     default:
@@ -1341,34 +2052,48 @@
   }
 }
 
-void ConnectionGraph::record_escape(Node *n, PhaseTransform *phase) {
-  if (_collecting)
-    record_escape_work(n, phase);
-}
-
 #ifndef PRODUCT
 void ConnectionGraph::dump() {
   PhaseGVN  *igvn = _compile->initial_gvn();
   bool first = true;
 
-  for (uint ni = 0; ni < (uint)_nodes->length(); ni++) {
-    PointsToNode *esp = _nodes->adr_at(ni);
-    if (esp->node_type() == PointsToNode::UnknownType || esp->_node == NULL)
+  uint size = (uint)_nodes->length();
+  for (uint ni = 0; ni < size; ni++) {
+    PointsToNode *ptn = _nodes->adr_at(ni);
+    PointsToNode::NodeType ptn_type = ptn->node_type();
+
+    if (ptn_type != PointsToNode::JavaObject || ptn->_node == NULL)
       continue;
-    PointsToNode::EscapeState es = escape_state(esp->_node, igvn);
-    if (es == PointsToNode::NoEscape || (Verbose &&
-            (es != PointsToNode::UnknownEscape || esp->edge_count() != 0))) {
-      // don't print null pointer node which almost every method has
-      if (esp->_node->Opcode() != Op_ConP || igvn->type(esp->_node) != TypePtr::NULL_PTR) {
-        if (first) {
-          tty->print("======== Connection graph for ");
-          C()->method()->print_short_name();
-          tty->cr();
-          first = false;
+    PointsToNode::EscapeState es = escape_state(ptn->_node, igvn);
+    if (ptn->_node->is_Allocate() && (es == PointsToNode::NoEscape || Verbose)) {
+      if (first) {
+        tty->cr();
+        tty->print("======== Connection graph for ");
+        C()->method()->print_short_name();
+        tty->cr();
+        first = false;
+      }
+      tty->print("%6d ", ni);
+      ptn->dump();
+      // Print all locals which reference this allocation
+      for (uint li = ni; li < size; li++) {
+        PointsToNode *ptn_loc = _nodes->adr_at(li);
+        PointsToNode::NodeType ptn_loc_type = ptn_loc->node_type();
+        if ( ptn_loc_type == PointsToNode::LocalVar && ptn_loc->_node != NULL &&
+             ptn_loc->edge_count() == 1 && ptn_loc->edge_target(0) == ni ) {
+          tty->print("%6d  LocalVar [[%d]]", li, ni);
+          _nodes->adr_at(li)->_node->dump();
         }
-        tty->print("%4d  ", ni);
-        esp->dump();
       }
+      if (Verbose) {
+        // Print all fields which reference this allocation
+        for (uint i = 0; i < ptn->edge_count(); i++) {
+          uint ei = ptn->edge_target(i);
+          tty->print("%6d  Field [[%d]]", ei, ni);
+          _nodes->adr_at(ei)->_node->dump();
+        }
+      }
+      tty->cr();
     }
   }
 }
--- a/src/share/vm/opto/escape.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/escape.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -25,14 +25,15 @@
 //
 // Adaptation for C2 of the escape analysis algorithm described in:
 //
-//     [Choi99] Jong-Deok Shoi, Manish Gupta, Mauricio Seffano, Vugranam C. Sreedhar,
-//              Sam Midkiff,  "Escape Analysis for Java", Procedings of ACM SIGPLAN
-//              OOPSLA  Conference, November 1, 1999
+// [Choi99] Jong-Deok Shoi, Manish Gupta, Mauricio Seffano,
+//          Vugranam C. Sreedhar, Sam Midkiff,
+//          "Escape Analysis for Java", Procedings of ACM SIGPLAN
+//          OOPSLA  Conference, November 1, 1999
 //
 // The flow-insensitive analysis described in the paper has been implemented.
 //
-// The analysis requires construction of a "connection graph" (CG) for the method being
-// analyzed.  The nodes of the connection graph are:
+// The analysis requires construction of a "connection graph" (CG) for
+// the method being analyzed.  The nodes of the connection graph are:
 //
 //     -  Java objects (JO)
 //     -  Local variables (LV)
@@ -40,47 +41,51 @@
 //
 // The CG contains 3 types of edges:
 //
-//   -  PointsTo  (-P>)     {LV,OF}  to JO
-//   -  Deferred  (-D>)    from {LV, OF}  to {LV, OF}
+//   -  PointsTo  (-P>)    {LV, OF} to JO
+//   -  Deferred  (-D>)    from {LV, OF} to {LV, OF}
 //   -  Field     (-F>)    from JO to OF
 //
 // The following  utility functions is used by the algorithm:
 //
-//   PointsTo(n)      - n is any CG node,  it returns the set of JO that n could
-//                      point to.
+//   PointsTo(n) - n is any CG node, it returns the set of JO that n could
+//                 point to.
 //
-// The algorithm describes how to construct the connection graph in the following 4 cases:
+// The algorithm describes how to construct the connection graph
+// in the following 4 cases:
 //
 //          Case                  Edges Created
 //
-// (1)   p   = new T()              LV  -P> JO
-// (2)   p   = q                    LV  -D> LV
-// (3)   p.f = q                    JO  -F> OF,  OF -D> LV
-// (4)   p   = q.f                  JO  -F> OF,  LV -D> OF
+// (1)   p   = new T()              LV -P> JO
+// (2)   p   = q                    LV -D> LV
+// (3)   p.f = q                    JO -F> OF,  OF -D> LV
+// (4)   p   = q.f                  JO -F> OF,  LV -D> OF
 //
-// In all these cases, p and q are local variables.  For static field references, we can
-// construct a local variable containing a reference to the static memory.
+// In all these cases, p and q are local variables.  For static field
+// references, we can construct a local variable containing a reference
+// to the static memory.
 //
 // C2 does not have local variables.  However for the purposes of constructing
 // the connection graph, the following IR nodes are treated as local variables:
 //     Phi    (pointer values)
 //     LoadP
-//     Proj  (value returned from callnodes including allocations)
-//     CheckCastPP
+//     Proj#5 (value returned from callnodes including allocations)
+//     CheckCastPP, CastPP
 //
-// The LoadP, Proj and CheckCastPP behave like variables assigned to only once.  Only
-// a Phi can have multiple assignments.  Each input to a Phi is treated
+// The LoadP, Proj and CheckCastPP behave like variables assigned to only once.
+// Only a Phi can have multiple assignments.  Each input to a Phi is treated
 // as an assignment to it.
 //
-// The following note types are JavaObject:
+// The following node types are JavaObject:
 //
 //     top()
 //     Allocate
 //     AllocateArray
 //     Parm  (for incoming arguments)
+//     CastX2P ("unsafe" operations)
 //     CreateEx
 //     ConP
 //     LoadKlass
+//     ThreadLocal
 //
 // AddP nodes are fields.
 //
@@ -89,7 +94,7 @@
 // source.  This results in a graph with no deferred edges, only:
 //
 //    LV -P> JO
-//    OF -P> JO
+//    OF -P> JO (the object whose oop is stored in the field)
 //    JO -F> OF
 //
 // Then, for each node which is GlobalEscape, anything it could point to
@@ -110,17 +115,18 @@
 friend class ConnectionGraph;
 public:
   typedef enum {
-    UnknownType    = 0,
-    JavaObject = 1,
-    LocalVar   = 2,
-    Field      = 3
+    UnknownType = 0,
+    JavaObject  = 1,
+    LocalVar    = 2,
+    Field       = 3
   } NodeType;
 
   typedef enum {
     UnknownEscape = 0,
-    NoEscape      = 1,
-    ArgEscape     = 2,
-    GlobalEscape  = 3
+    NoEscape      = 1, // A scalar replaceable object with unique type.
+    ArgEscape     = 2, // An object passed as argument or referenced by
+                       // argument (and not globally escape during call).
+    GlobalEscape  = 3  // An object escapes the method and thread.
   } EscapeState;
 
   typedef enum {
@@ -140,18 +146,24 @@
 
   NodeType             _type;
   EscapeState          _escape;
-  GrowableArray<uint>* _edges;  // outgoing edges
-  int                  _offset; // for fields
+  GrowableArray<uint>* _edges;   // outgoing edges
 
-  bool       _unique_type;       // For allocated objects, this node may be a unique type
 public:
-  Node*      _node;              // Ideal node corresponding to this PointsTo node
-  int        _inputs_processed;  // the number of Phi inputs that have been processed so far
-  bool       _hidden_alias;      // this node is an argument to a function which may return it
-                                 // creating a hidden alias
+  Node* _node;              // Ideal node corresponding to this PointsTo node.
+  int   _offset;            // Object fields offsets.
+  bool  _scalar_replaceable;// Not escaped object could be replaced with scalar
+  bool  _hidden_alias;      // This node is an argument to a function.
+                            // which may return it creating a hidden alias.
 
+  PointsToNode():
+    _type(UnknownType),
+    _escape(UnknownEscape),
+    _edges(NULL),
+    _node(NULL),
+    _offset(-1),
+    _scalar_replaceable(true),
+    _hidden_alias(false) {}
 
-  PointsToNode(): _offset(-1), _type(UnknownType), _escape(UnknownEscape), _edges(NULL), _node(NULL), _inputs_processed(0), _hidden_alias(false), _unique_type(true) {}
 
   EscapeState escape_state() const { return _escape; }
   NodeType node_type() const { return _type;}
@@ -182,22 +194,28 @@
 
 class ConnectionGraph: public ResourceObj {
 private:
-  enum {
-    INITIAL_NODE_COUNT = 100                    // initial size of _nodes array
-  };
+  GrowableArray<PointsToNode>* _nodes; // Connection graph nodes indexed
+                                       // by ideal node index.
 
+  Unique_Node_List  _delayed_worklist; // Nodes to be processed before
+                                       // the call build_connection_graph().
+
+  VectorSet                _processed; // Records which nodes have been
+                                       // processed.
 
-  GrowableArray<PointsToNode>* _nodes;          // connection graph nodes  Indexed by ideal
-                                                // node index
-  Unique_Node_List             _deferred;       // Phi's to be processed after parsing
-  VectorSet                    _processed;      // records which nodes have been processed
-  bool                         _collecting;     // indicates whether escape information is
-                                                // still being collected.  If false, no new
-                                                // nodes will be processed
-  uint                         _phantom_object; // index of globally escaping object that
-                                                // pointer values loaded from a field which
-                                                // has not been set are assumed to point to
-  Compile *                    _compile;        // Compile object for current compilation
+  bool                    _collecting; // Indicates whether escape information
+                                       // is still being collected. If false,
+                                       // no new nodes will be processed.
+
+  bool               _has_allocations; // Indicates whether method has any
+                                       // non-escaping allocations.
+
+  uint                _phantom_object; // Index of globally escaping object
+                                       // that pointer values loaded from
+                                       // a field which has not been set
+                                       // are assumed to point to.
+
+  Compile *                  _compile; // Compile object for current compilation
 
   // address of an element in _nodes.  Used when the element is to be modified
   PointsToNode *ptnode_adr(uint idx) {
@@ -208,8 +226,11 @@
     return _nodes->adr_at(idx);
   }
 
+  // Add node to ConnectionGraph.
+  void add_node(Node *n, PointsToNode::NodeType nt, PointsToNode::EscapeState es, bool done);
+
   // offset of a field reference
-  int type_to_offset(const Type *t);
+  int address_offset(Node* adr, PhaseTransform *phase);
 
   // compute the escape state for arguments to a call
   void process_call_arguments(CallNode *call, PhaseTransform *phase);
@@ -217,12 +238,11 @@
   // compute the escape state for the return value of a call
   void process_call_result(ProjNode *resproj, PhaseTransform *phase);
 
-  // compute the escape state of a Phi.  This may be called multiple
-  // times as new inputs are added to the Phi.
-  void process_phi_escape(PhiNode *phi, PhaseTransform *phase);
+  // Populate Connection Graph with Ideal nodes.
+  void record_for_escape_analysis(Node *n, PhaseTransform *phase);
 
-  // compute the escape state of an ideal node.
-  void record_escape_work(Node *n, PhaseTransform *phase);
+  // Build Connection Graph and set nodes escape state.
+  void build_connection_graph(Node *n, PhaseTransform *phase);
 
   // walk the connection graph starting at the node corresponding to "n" and
   // add the index of everything it could point to, to "ptset".  This may cause
@@ -241,8 +261,8 @@
   // a pointsto edge is added if it is a JavaObject
   void add_edge_from_fields(uint adr, uint to_i, int offs);
 
-  // Add a deferred  edge from node given by "from_i" to any field of adr_i whose offset
-  // matches "offset"
+  // Add a deferred  edge from node given by "from_i" to any field
+  // of adr_i whose offset matches "offset"
   void add_deferred_edge_to_fields(uint from_i, uint adr, int offs);
 
 
@@ -262,6 +282,8 @@
   PhiNode *create_split_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *>  &orig_phi_worklist, PhaseGVN  *igvn, bool &new_created);
   PhiNode *split_memory_phi(PhiNode *orig_phi, int alias_idx, GrowableArray<PhiNode *>  &orig_phi_worklist, PhaseGVN  *igvn);
   Node *find_mem(Node *mem, int alias_idx, PhaseGVN  *igvn);
+  Node *find_inst_mem(Node *mem, int alias_idx,GrowableArray<PhiNode *>  &orig_phi_worklist,  PhaseGVN  *igvn);
+
   // Propagate unique types created for unescaped allocated objects
   // through the graph
   void split_unique_types(GrowableArray<Node *>  &alloc_worklist);
@@ -285,26 +307,24 @@
   // Set the escape state of a node
   void set_escape_state(uint ni, PointsToNode::EscapeState es);
 
-  // bypass any casts and return the node they refer to
-  Node * skip_casts(Node *n);
-
   // Get Compile object for current compilation.
   Compile *C() const        { return _compile; }
 
 public:
   ConnectionGraph(Compile *C);
 
-  // record a Phi for later processing.
-  void record_for_escape_analysis(Node *n);
-
-  // process a node and  fill in its connection graph node
-  void record_escape(Node *n, PhaseTransform *phase);
-
-  // All nodes have been recorded, compute the escape information
+  // Compute the escape information
   void compute_escape();
 
   // escape state of a node
   PointsToNode::EscapeState escape_state(Node *n, PhaseTransform *phase);
+  // other information we have collected
+  bool is_scalar_replaceable(Node *n) {
+    if (_collecting)
+      return false;
+    PointsToNode  ptn = _nodes->at_grow(n->_idx);
+    return ptn.escape_state() == PointsToNode::NoEscape && ptn._scalar_replaceable;
+  }
 
   bool hidden_alias(Node *n) {
     if (_collecting)
--- a/src/share/vm/opto/graphKit.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/graphKit.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -857,6 +857,13 @@
     for (j = 0; j < l; j++)
       call->set_req(p++, in_map->in(k+j));
 
+    // Copy any scalar object fields.
+    k = in_jvms->scloff();
+    l = in_jvms->scl_size();
+    out_jvms->set_scloff(p);
+    for (j = 0; j < l; j++)
+      call->set_req(p++, in_map->in(k+j));
+
     // Finish the new jvms.
     out_jvms->set_endoff(p);
 
@@ -864,6 +871,7 @@
     assert(out_jvms->depth()      == in_jvms->depth(),      "depth must match");
     assert(out_jvms->loc_size()   == in_jvms->loc_size(),   "size must match");
     assert(out_jvms->mon_size()   == in_jvms->mon_size(),   "size must match");
+    assert(out_jvms->scl_size()   == in_jvms->scl_size(),   "size must match");
     assert(out_jvms->debug_size() == in_jvms->debug_size(), "size must match");
 
     // Update the two tail pointers in parallel.
@@ -2914,10 +2922,22 @@
   const TypeOopPtr* oop_type = tklass->as_instance_type();
 
   // Now generate allocation code
+
+  // With escape analysis, the entire memory state is needed to be able to
+  // eliminate the allocation.  If the allocations cannot be eliminated, this
+  // will be optimized to the raw slice when the allocation is expanded.
+  Node *mem;
+  if (C->do_escape_analysis()) {
+    mem = reset_memory();
+    set_all_memory(mem);
+  } else {
+    mem = memory(Compile::AliasIdxRaw);
+  }
+
   AllocateNode* alloc
     = new (C, AllocateNode::ParmLimit)
         AllocateNode(C, AllocateNode::alloc_type(),
-                     control(), memory(Compile::AliasIdxRaw), i_o(),
+                     control(), mem, i_o(),
                      size, klass_node,
                      initial_slow_test);
 
@@ -3048,11 +3068,23 @@
   }
 
   // Now generate allocation code
+
+  // With escape analysis, the entire memory state is needed to be able to
+  // eliminate the allocation.  If the allocations cannot be eliminated, this
+  // will be optimized to the raw slice when the allocation is expanded.
+  Node *mem;
+  if (C->do_escape_analysis()) {
+    mem = reset_memory();
+    set_all_memory(mem);
+  } else {
+    mem = memory(Compile::AliasIdxRaw);
+  }
+
   // Create the AllocateArrayNode and its result projections
   AllocateArrayNode* alloc
     = new (C, AllocateArrayNode::ParmLimit)
         AllocateArrayNode(C, AllocateArrayNode::alloc_type(),
-                          control(), memory(Compile::AliasIdxRaw), i_o(),
+                          control(), mem, i_o(),
                           size, klass_node,
                           initial_slow_test,
                           length);
--- a/src/share/vm/opto/locknode.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/locknode.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -36,7 +36,8 @@
 
 uint BoxLockNode::size_of() const { return sizeof(*this); }
 
-BoxLockNode::BoxLockNode( int slot ) : Node( Compile::current()->root() ), _slot(slot) {
+BoxLockNode::BoxLockNode( int slot ) : Node( Compile::current()->root() ),
+                                       _slot(slot), _is_eliminated(false) {
   init_class_id(Class_BoxLock);
   init_flags(Flag_rematerialize);
   OptoReg::Name reg = OptoReg::stack2reg(_slot);
--- a/src/share/vm/opto/locknode.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/locknode.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -27,6 +27,7 @@
 public:
   const int _slot;
   RegMask   _inmask;
+  bool _is_eliminated;    // indicates this lock was safely eliminated
 
   BoxLockNode( int lock );
   virtual int Opcode() const;
@@ -42,6 +43,10 @@
 
   static OptoReg::Name stack_slot(Node* box_node);
 
+  bool is_eliminated()  { return _is_eliminated; }
+  // mark lock as eliminated.
+  void set_eliminated() { _is_eliminated = true; }
+
 #ifndef PRODUCT
   virtual void format( PhaseRegAlloc *, outputStream *st ) const;
   virtual void dump_spec(outputStream *st) const { st->print("  Lock %d",_slot); }
--- a/src/share/vm/opto/loopopts.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/loopopts.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -29,10 +29,26 @@
 //------------------------------split_thru_phi---------------------------------
 // Split Node 'n' through merge point if there is enough win.
 Node *PhaseIdealLoop::split_thru_phi( Node *n, Node *region, int policy ) {
+  if (n->Opcode() == Op_ConvI2L && n->bottom_type() != TypeLong::LONG) {
+    // ConvI2L may have type information on it which is unsafe to push up
+    // so disable this for now
+    return NULL;
+  }
   int wins = 0;
   assert( !n->is_CFG(), "" );
   assert( region->is_Region(), "" );
-  Node *phi = new (C, region->req()) PhiNode( region, n->bottom_type() );
+
+  const Type* type = n->bottom_type();
+  const TypeOopPtr *t_oop = _igvn.type(n)->isa_oopptr();
+  Node *phi;
+  if( t_oop != NULL && t_oop->is_instance_field() ) {
+    int iid    = t_oop->instance_id();
+    int index  = C->get_alias_index(t_oop);
+    int offset = t_oop->offset();
+    phi = new (C,region->req()) PhiNode(region, type, NULL, iid, index, offset);
+  } else {
+    phi = new (C,region->req()) PhiNode(region, type);
+  }
   uint old_unique = C->unique();
   for( uint i = 1; i < region->req(); i++ ) {
     Node *x;
--- a/src/share/vm/opto/macro.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/macro.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -54,15 +54,30 @@
   uint new_dbg_start = newcall->tf()->domain()->cnt();
   int jvms_adj  = new_dbg_start - old_dbg_start;
   assert (new_dbg_start == newcall->req(), "argument count mismatch");
+
+  Dict* sosn_map = new Dict(cmpkey,hashkey);
   for (uint i = old_dbg_start; i < oldcall->req(); i++) {
-    newcall->add_req(oldcall->in(i));
+    Node* old_in = oldcall->in(i);
+    // Clone old SafePointScalarObjectNodes, adjusting their field contents.
+    if (old_in->is_SafePointScalarObject()) {
+      SafePointScalarObjectNode* old_sosn = old_in->as_SafePointScalarObject();
+      uint old_unique = C->unique();
+      Node* new_in = old_sosn->clone(jvms_adj, sosn_map);
+      if (old_unique != C->unique()) {
+        new_in = transform_later(new_in); // Register new node.
+      }
+      old_in = new_in;
+    }
+    newcall->add_req(old_in);
   }
+
   newcall->set_jvms(oldcall->jvms());
   for (JVMState *jvms = newcall->jvms(); jvms != NULL; jvms = jvms->caller()) {
     jvms->set_map(newcall);
     jvms->set_locoff(jvms->locoff()+jvms_adj);
     jvms->set_stkoff(jvms->stkoff()+jvms_adj);
     jvms->set_monoff(jvms->monoff()+jvms_adj);
+    jvms->set_scloff(jvms->scloff()+jvms_adj);
     jvms->set_endoff(jvms->endoff()+jvms_adj);
   }
 }
@@ -166,6 +181,622 @@
 
 }
 
+// Eliminate a card mark sequence.  p2x is a ConvP2XNode
+void PhaseMacroExpand::eliminate_card_mark(Node *p2x) {
+  assert(p2x->Opcode() == Op_CastP2X, "ConvP2XNode required");
+  Node *shift = p2x->unique_out();
+  Node *addp = shift->unique_out();
+  for (DUIterator_Last jmin, j = addp->last_outs(jmin); j >= jmin; --j) {
+    Node *st = addp->last_out(j);
+    assert(st->is_Store(), "store required");
+    _igvn.replace_node(st, st->in(MemNode::Memory));
+  }
+}
+
+// Search for a memory operation for the specified memory slice.
+static Node *scan_mem_chain(Node *mem, int alias_idx, int offset, Node *start_mem, Node *alloc) {
+  Node *orig_mem = mem;
+  Node *alloc_mem = alloc->in(TypeFunc::Memory);
+  while (true) {
+    if (mem == alloc_mem || mem == start_mem ) {
+      return mem;  // hit one of our sentinals
+    } else if (mem->is_MergeMem()) {
+      mem = mem->as_MergeMem()->memory_at(alias_idx);
+    } else if (mem->is_Proj() && mem->as_Proj()->_con == TypeFunc::Memory) {
+      Node *in = mem->in(0);
+      // we can safely skip over safepoints, calls, locks and membars because we
+      // already know that the object is safe to eliminate.
+      if (in->is_Initialize() && in->as_Initialize()->allocation() == alloc) {
+        return in;
+      } else if (in->is_Call() || in->is_MemBar()) {
+        mem = in->in(TypeFunc::Memory);
+      } else {
+        assert(false, "unexpected projection");
+      }
+    } else if (mem->is_Store()) {
+      const TypePtr* atype = mem->as_Store()->adr_type();
+      int adr_idx = Compile::current()->get_alias_index(atype);
+      if (adr_idx == alias_idx) {
+        assert(atype->isa_oopptr(), "address type must be oopptr");
+        int adr_offset = atype->offset();
+        uint adr_iid = atype->is_oopptr()->instance_id();
+        // Array elements references have the same alias_idx
+        // but different offset and different instance_id.
+        if (adr_offset == offset && adr_iid == alloc->_idx)
+          return mem;
+      } else {
+        assert(adr_idx == Compile::AliasIdxRaw, "address must match or be raw");
+      }
+      mem = mem->in(MemNode::Memory);
+    } else {
+      return mem;
+    }
+    if (mem == orig_mem)
+      return mem;
+  }
+}
+
+//
+// Given a Memory Phi, compute a value Phi containing the values from stores
+// on the input paths.
+// Note: this function is recursive, its depth is limied by the "level" argument
+// Returns the computed Phi, or NULL if it cannot compute it.
+Node *PhaseMacroExpand::value_from_mem_phi(Node *mem, BasicType ft, const Type *phi_type, const TypeOopPtr *adr_t, Node *alloc, int level) {
+
+  if (level <= 0) {
+    return NULL;
+  }
+  int alias_idx = C->get_alias_index(adr_t);
+  int offset = adr_t->offset();
+  int instance_id = adr_t->instance_id();
+
+  Node *start_mem = C->start()->proj_out(TypeFunc::Memory);
+  Node *alloc_mem = alloc->in(TypeFunc::Memory);
+
+  uint length = mem->req();
+  GrowableArray <Node *> values(length, length, NULL);
+
+  for (uint j = 1; j < length; j++) {
+    Node *in = mem->in(j);
+    if (in == NULL || in->is_top()) {
+      values.at_put(j, in);
+    } else  {
+      Node *val = scan_mem_chain(in, alias_idx, offset, start_mem, alloc);
+      if (val == start_mem || val == alloc_mem) {
+        // hit a sentinel, return appropriate 0 value
+        values.at_put(j, _igvn.zerocon(ft));
+        continue;
+      }
+      if (val->is_Initialize()) {
+        val = val->as_Initialize()->find_captured_store(offset, type2aelembytes(ft), &_igvn);
+      }
+      if (val == NULL) {
+        return NULL;  // can't find a value on this path
+      }
+      if (val == mem) {
+        values.at_put(j, mem);
+      } else if (val->is_Store()) {
+        values.at_put(j, val->in(MemNode::ValueIn));
+      } else if(val->is_Proj() && val->in(0) == alloc) {
+        values.at_put(j, _igvn.zerocon(ft));
+      } else if (val->is_Phi()) {
+        // Check if an appropriate node already exists.
+        Node* region = val->in(0);
+        Node* old_phi = NULL;
+        for (DUIterator_Fast kmax, k = region->fast_outs(kmax); k < kmax; k++) {
+          Node* phi = region->fast_out(k);
+          if (phi->is_Phi() && phi != val &&
+              phi->as_Phi()->is_same_inst_field(phi_type, instance_id, alias_idx, offset)) {
+            old_phi = phi;
+            break;
+          }
+        }
+        if (old_phi == NULL) {
+          val = value_from_mem_phi(val, ft, phi_type, adr_t, alloc, level-1);
+          if (val == NULL) {
+            return NULL;
+          }
+          values.at_put(j, val);
+        } else {
+          values.at_put(j, old_phi);
+        }
+      } else {
+        return NULL;  // unknown node  on this path
+      }
+    }
+  }
+  // create a new Phi for the value
+  PhiNode *phi = new (C, length) PhiNode(mem->in(0), phi_type, NULL, instance_id, alias_idx, offset);
+  for (uint j = 1; j < length; j++) {
+    if (values.at(j) == mem) {
+      phi->init_req(j, phi);
+    } else {
+      phi->init_req(j, values.at(j));
+    }
+  }
+  transform_later(phi);
+  return phi;
+}
+
+// Search the last value stored into the object's field.
+Node *PhaseMacroExpand::value_from_mem(Node *sfpt_mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc) {
+  assert(adr_t->is_instance_field(), "instance required");
+  uint instance_id = adr_t->instance_id();
+  assert(instance_id == alloc->_idx, "wrong allocation");
+
+  int alias_idx = C->get_alias_index(adr_t);
+  int offset = adr_t->offset();
+  Node *start_mem = C->start()->proj_out(TypeFunc::Memory);
+  Node *alloc_ctrl = alloc->in(TypeFunc::Control);
+  Node *alloc_mem = alloc->in(TypeFunc::Memory);
+  VectorSet visited(Thread::current()->resource_area());
+
+
+  bool done = sfpt_mem == alloc_mem;
+  Node *mem = sfpt_mem;
+  while (!done) {
+    if (visited.test_set(mem->_idx)) {
+      return NULL;  // found a loop, give up
+    }
+    mem = scan_mem_chain(mem, alias_idx, offset, start_mem, alloc);
+    if (mem == start_mem || mem == alloc_mem) {
+      done = true;  // hit a sentinel, return appropriate 0 value
+    } else if (mem->is_Initialize()) {
+      mem = mem->as_Initialize()->find_captured_store(offset, type2aelembytes(ft), &_igvn);
+      if (mem == NULL) {
+        done = true; // Something go wrong.
+      } else if (mem->is_Store()) {
+        const TypePtr* atype = mem->as_Store()->adr_type();
+        assert(C->get_alias_index(atype) == Compile::AliasIdxRaw, "store is correct memory slice");
+        done = true;
+      }
+    } else if (mem->is_Store()) {
+      const TypeOopPtr* atype = mem->as_Store()->adr_type()->isa_oopptr();
+      assert(atype != NULL, "address type must be oopptr");
+      assert(C->get_alias_index(atype) == alias_idx &&
+             atype->is_instance_field() && atype->offset() == offset &&
+             atype->instance_id() == instance_id, "store is correct memory slice");
+      done = true;
+    } else if (mem->is_Phi()) {
+      // try to find a phi's unique input
+      Node *unique_input = NULL;
+      Node *top = C->top();
+      for (uint i = 1; i < mem->req(); i++) {
+        Node *n = scan_mem_chain(mem->in(i), alias_idx, offset, start_mem, alloc);
+        if (n == NULL || n == top || n == mem) {
+          continue;
+        } else if (unique_input == NULL) {
+          unique_input = n;
+        } else if (unique_input != n) {
+          unique_input = top;
+          break;
+        }
+      }
+      if (unique_input != NULL && unique_input != top) {
+        mem = unique_input;
+      } else {
+        done = true;
+      }
+    } else {
+      assert(false, "unexpected node");
+    }
+  }
+  if (mem != NULL) {
+    if (mem == start_mem || mem == alloc_mem) {
+      // hit a sentinel, return appropriate 0 value
+      return _igvn.zerocon(ft);
+    } else if (mem->is_Store()) {
+      return mem->in(MemNode::ValueIn);
+    } else if (mem->is_Phi()) {
+      // attempt to produce a Phi reflecting the values on the input paths of the Phi
+      Node * phi = value_from_mem_phi(mem, ft, ftype, adr_t, alloc, 8);
+      if (phi != NULL) {
+        return phi;
+      }
+    }
+  }
+  // Something go wrong.
+  return NULL;
+}
+
+// Check the possibility of scalar replacement.
+bool PhaseMacroExpand::can_eliminate_allocation(AllocateNode *alloc, GrowableArray <SafePointNode *>& safepoints) {
+  //  Scan the uses of the allocation to check for anything that would
+  //  prevent us from eliminating it.
+  NOT_PRODUCT( const char* fail_eliminate = NULL; )
+  DEBUG_ONLY( Node* disq_node = NULL; )
+  bool  can_eliminate = true;
+
+  Node* res = alloc->result_cast();
+  const TypeOopPtr* res_type = NULL;
+  if (res == NULL) {
+    // All users were eliminated.
+  } else if (!res->is_CheckCastPP()) {
+    alloc->_is_scalar_replaceable = false;  // don't try again
+    NOT_PRODUCT(fail_eliminate = "Allocation does not have unique CheckCastPP";)
+    can_eliminate = false;
+  } else {
+    res_type = _igvn.type(res)->isa_oopptr();
+    if (res_type == NULL) {
+      NOT_PRODUCT(fail_eliminate = "Neither instance or array allocation";)
+      can_eliminate = false;
+    } else if (res_type->isa_aryptr()) {
+      int length = alloc->in(AllocateNode::ALength)->find_int_con(-1);
+      if (length < 0) {
+        NOT_PRODUCT(fail_eliminate = "Array's size is not constant";)
+        can_eliminate = false;
+      }
+    }
+  }
+
+  if (can_eliminate && res != NULL) {
+    for (DUIterator_Fast jmax, j = res->fast_outs(jmax);
+                               j < jmax && can_eliminate; j++) {
+      Node* use = res->fast_out(j);
+
+      if (use->is_AddP()) {
+        const TypePtr* addp_type = _igvn.type(use)->is_ptr();
+        int offset = addp_type->offset();
+
+        if (offset == Type::OffsetTop || offset == Type::OffsetBot) {
+          NOT_PRODUCT(fail_eliminate = "Undefined field referrence";)
+          can_eliminate = false;
+          break;
+        }
+        for (DUIterator_Fast kmax, k = use->fast_outs(kmax);
+                                   k < kmax && can_eliminate; k++) {
+          Node* n = use->fast_out(k);
+          if (!n->is_Store() && n->Opcode() != Op_CastP2X) {
+            DEBUG_ONLY(disq_node = n;)
+            if (n->is_Load()) {
+              NOT_PRODUCT(fail_eliminate = "Field load";)
+            } else {
+              NOT_PRODUCT(fail_eliminate = "Not store field referrence";)
+            }
+            can_eliminate = false;
+          }
+        }
+      } else if (use->is_SafePoint()) {
+        SafePointNode* sfpt = use->as_SafePoint();
+        if (sfpt->has_non_debug_use(res)) {
+          // Object is passed as argument.
+          DEBUG_ONLY(disq_node = use;)
+          NOT_PRODUCT(fail_eliminate = "Object is passed as argument";)
+          can_eliminate = false;
+        }
+        Node* sfptMem = sfpt->memory();
+        if (sfptMem == NULL || sfptMem->is_top()) {
+          DEBUG_ONLY(disq_node = use;)
+          NOT_PRODUCT(fail_eliminate = "NULL or TOP memory";)
+          can_eliminate = false;
+        } else {
+          safepoints.append_if_missing(sfpt);
+        }
+      } else if (use->Opcode() != Op_CastP2X) { // CastP2X is used by card mark
+        if (use->is_Phi()) {
+          if (use->outcnt() == 1 && use->unique_out()->Opcode() == Op_Return) {
+            NOT_PRODUCT(fail_eliminate = "Object is return value";)
+          } else {
+            NOT_PRODUCT(fail_eliminate = "Object is referenced by Phi";)
+          }
+          DEBUG_ONLY(disq_node = use;)
+        } else {
+          if (use->Opcode() == Op_Return) {
+            NOT_PRODUCT(fail_eliminate = "Object is return value";)
+          }else {
+            NOT_PRODUCT(fail_eliminate = "Object is referenced by node";)
+          }
+          DEBUG_ONLY(disq_node = use;)
+        }
+        can_eliminate = false;
+      }
+    }
+  }
+
+#ifndef PRODUCT
+  if (PrintEliminateAllocations) {
+    if (can_eliminate) {
+      tty->print("Scalar ");
+      if (res == NULL)
+        alloc->dump();
+      else
+        res->dump();
+    } else {
+      tty->print("NotScalar (%s)", fail_eliminate);
+      if (res == NULL)
+        alloc->dump();
+      else
+        res->dump();
+#ifdef ASSERT
+      if (disq_node != NULL) {
+          tty->print("  >>>> ");
+          disq_node->dump();
+      }
+#endif /*ASSERT*/
+    }
+  }
+#endif
+  return can_eliminate;
+}
+
+// Do scalar replacement.
+bool PhaseMacroExpand::scalar_replacement(AllocateNode *alloc, GrowableArray <SafePointNode *>& safepoints) {
+  GrowableArray <SafePointNode *> safepoints_done;
+
+  ciKlass* klass = NULL;
+  ciInstanceKlass* iklass = NULL;
+  int nfields = 0;
+  int array_base;
+  int element_size;
+  BasicType basic_elem_type;
+  ciType* elem_type;
+
+  Node* res = alloc->result_cast();
+  const TypeOopPtr* res_type = NULL;
+  if (res != NULL) { // Could be NULL when there are no users
+    res_type = _igvn.type(res)->isa_oopptr();
+  }
+
+  if (res != NULL) {
+    klass = res_type->klass();
+    if (res_type->isa_instptr()) {
+      // find the fields of the class which will be needed for safepoint debug information
+      assert(klass->is_instance_klass(), "must be an instance klass.");
+      iklass = klass->as_instance_klass();
+      nfields = iklass->nof_nonstatic_fields();
+    } else {
+      // find the array's elements which will be needed for safepoint debug information
+      nfields = alloc->in(AllocateNode::ALength)->find_int_con(-1);
+      assert(klass->is_array_klass() && nfields >= 0, "must be an array klass.");
+      elem_type = klass->as_array_klass()->element_type();
+      basic_elem_type = elem_type->basic_type();
+      array_base = arrayOopDesc::base_offset_in_bytes(basic_elem_type);
+      element_size = type2aelembytes(basic_elem_type);
+    }
+  }
+  //
+  // Process the safepoint uses
+  //
+  while (safepoints.length() > 0) {
+    SafePointNode* sfpt = safepoints.pop();
+    Node* mem = sfpt->memory();
+    uint first_ind = sfpt->req();
+    SafePointScalarObjectNode* sobj = new (C, 1) SafePointScalarObjectNode(res_type,
+#ifdef ASSERT
+                                                 alloc,
+#endif
+                                                 first_ind, nfields);
+    sobj->init_req(0, sfpt->in(TypeFunc::Control));
+    transform_later(sobj);
+
+    // Scan object's fields adding an input to the safepoint for each field.
+    for (int j = 0; j < nfields; j++) {
+      int offset;
+      ciField* field = NULL;
+      if (iklass != NULL) {
+        field = iklass->nonstatic_field_at(j);
+        offset = field->offset();
+        elem_type = field->type();
+        basic_elem_type = field->layout_type();
+      } else {
+        offset = array_base + j * element_size;
+      }
+
+      const Type *field_type;
+      // The next code is taken from Parse::do_get_xxx().
+      if (basic_elem_type == T_OBJECT) {
+        if (!elem_type->is_loaded()) {
+          field_type = TypeInstPtr::BOTTOM;
+        } else if (field != NULL && field->is_constant()) {
+          // This can happen if the constant oop is non-perm.
+          ciObject* con = field->constant_value().as_object();
+          // Do not "join" in the previous type; it doesn't add value,
+          // and may yield a vacuous result if the field is of interface type.
+          field_type = TypeOopPtr::make_from_constant(con)->isa_oopptr();
+          assert(field_type != NULL, "field singleton type must be consistent");
+        } else {
+          field_type = TypeOopPtr::make_from_klass(elem_type->as_klass());
+        }
+      } else {
+        field_type = Type::get_const_basic_type(basic_elem_type);
+      }
+
+      const TypeOopPtr *field_addr_type = res_type->add_offset(offset)->isa_oopptr();
+
+      Node *field_val = value_from_mem(mem, basic_elem_type, field_type, field_addr_type, alloc);
+      if (field_val == NULL) {
+        // we weren't able to find a value for this field,
+        // give up on eliminating this allocation
+        alloc->_is_scalar_replaceable = false;  // don't try again
+        // remove any extra entries we added to the safepoint
+        uint last = sfpt->req() - 1;
+        for (int k = 0;  k < j; k++) {
+          sfpt->del_req(last--);
+        }
+        // rollback processed safepoints
+        while (safepoints_done.length() > 0) {
+          SafePointNode* sfpt_done = safepoints_done.pop();
+          // remove any extra entries we added to the safepoint
+          last = sfpt_done->req() - 1;
+          for (int k = 0;  k < nfields; k++) {
+            sfpt_done->del_req(last--);
+          }
+          JVMState *jvms = sfpt_done->jvms();
+          jvms->set_endoff(sfpt_done->req());
+          // Now make a pass over the debug information replacing any references
+          // to SafePointScalarObjectNode with the allocated object.
+          int start = jvms->debug_start();
+          int end   = jvms->debug_end();
+          for (int i = start; i < end; i++) {
+            if (sfpt_done->in(i)->is_SafePointScalarObject()) {
+              SafePointScalarObjectNode* scobj = sfpt_done->in(i)->as_SafePointScalarObject();
+              if (scobj->first_index() == sfpt_done->req() &&
+                  scobj->n_fields() == (uint)nfields) {
+                assert(scobj->alloc() == alloc, "sanity");
+                sfpt_done->set_req(i, res);
+              }
+            }
+          }
+        }
+#ifndef PRODUCT
+        if (PrintEliminateAllocations) {
+          if (field != NULL) {
+            tty->print("=== At SafePoint node %d can't find value of Field: ",
+                       sfpt->_idx);
+            field->print();
+            int field_idx = C->get_alias_index(field_addr_type);
+            tty->print(" (alias_idx=%d)", field_idx);
+          } else { // Array's element
+            tty->print("=== At SafePoint node %d can't find value of array element [%d]",
+                       sfpt->_idx, j);
+          }
+          tty->print(", which prevents elimination of: ");
+          if (res == NULL)
+            alloc->dump();
+          else
+            res->dump();
+        }
+#endif
+        return false;
+      }
+      sfpt->add_req(field_val);
+    }
+    JVMState *jvms = sfpt->jvms();
+    jvms->set_endoff(sfpt->req());
+    // Now make a pass over the debug information replacing any references
+    // to the allocated object with "sobj"
+    int start = jvms->debug_start();
+    int end   = jvms->debug_end();
+    for (int i = start; i < end; i++) {
+      if (sfpt->in(i) == res) {
+        sfpt->set_req(i, sobj);
+      }
+    }
+    safepoints_done.append_if_missing(sfpt); // keep it for rollback
+  }
+  return true;
+}
+
+// Process users of eliminated allocation.
+void PhaseMacroExpand::process_users_of_allocation(AllocateNode *alloc) {
+  Node* res = alloc->result_cast();
+  if (res != NULL) {
+    for (DUIterator_Last jmin, j = res->last_outs(jmin); j >= jmin; ) {
+      Node *use = res->last_out(j);
+      uint oc1 = res->outcnt();
+
+      if (use->is_AddP()) {
+        for (DUIterator_Last kmin, k = use->last_outs(kmin); k >= kmin; ) {
+          Node *n = use->last_out(k);
+          uint oc2 = use->outcnt();
+          if (n->is_Store()) {
+            _igvn.replace_node(n, n->in(MemNode::Memory));
+          } else {
+            assert( n->Opcode() == Op_CastP2X, "CastP2X required");
+            eliminate_card_mark(n);
+          }
+          k -= (oc2 - use->outcnt());
+        }
+      } else {
+        assert( !use->is_SafePoint(), "safepoint uses must have been already elimiated");
+        assert( use->Opcode() == Op_CastP2X, "CastP2X required");
+        eliminate_card_mark(use);
+      }
+      j -= (oc1 - res->outcnt());
+    }
+    assert(res->outcnt() == 0, "all uses of allocated objects must be deleted");
+    _igvn.remove_dead_node(res);
+  }
+
+  //
+  // Process other users of allocation's projections
+  //
+  if (_resproj != NULL && _resproj->outcnt() != 0) {
+    for (DUIterator_Last jmin, j = _resproj->last_outs(jmin); j >= jmin; ) {
+      Node *use = _resproj->last_out(j);
+      uint oc1 = _resproj->outcnt();
+      if (use->is_Initialize()) {
+        // Eliminate Initialize node.
+        InitializeNode *init = use->as_Initialize();
+        assert(init->outcnt() <= 2, "only a control and memory projection expected");
+        Node *ctrl_proj = init->proj_out(TypeFunc::Control);
+        if (ctrl_proj != NULL) {
+           assert(init->in(TypeFunc::Control) == _fallthroughcatchproj, "allocation control projection");
+          _igvn.replace_node(ctrl_proj, _fallthroughcatchproj);
+        }
+        Node *mem_proj = init->proj_out(TypeFunc::Memory);
+        if (mem_proj != NULL) {
+          Node *mem = init->in(TypeFunc::Memory);
+#ifdef ASSERT
+          if (mem->is_MergeMem()) {
+            assert(mem->in(TypeFunc::Memory) == _memproj_fallthrough, "allocation memory projection");
+          } else {
+            assert(mem == _memproj_fallthrough, "allocation memory projection");
+          }
+#endif
+          _igvn.replace_node(mem_proj, mem);
+        }
+      } else if (use->is_AddP()) {
+        // raw memory addresses used only by the initialization
+        _igvn.hash_delete(use);
+        _igvn.subsume_node(use, C->top());
+      } else  {
+        assert(false, "only Initialize or AddP expected");
+      }
+      j -= (oc1 - _resproj->outcnt());
+    }
+  }
+  if (_fallthroughcatchproj != NULL) {
+    _igvn.replace_node(_fallthroughcatchproj, alloc->in(TypeFunc::Control));
+  }
+  if (_memproj_fallthrough != NULL) {
+    _igvn.replace_node(_memproj_fallthrough, alloc->in(TypeFunc::Memory));
+  }
+  if (_memproj_catchall != NULL) {
+    _igvn.replace_node(_memproj_catchall, C->top());
+  }
+  if (_ioproj_fallthrough != NULL) {
+    _igvn.replace_node(_ioproj_fallthrough, alloc->in(TypeFunc::I_O));
+  }
+  if (_ioproj_catchall != NULL) {
+    _igvn.replace_node(_ioproj_catchall, C->top());
+  }
+  if (_catchallcatchproj != NULL) {
+    _igvn.replace_node(_catchallcatchproj, C->top());
+  }
+}
+
+bool PhaseMacroExpand::eliminate_allocate_node(AllocateNode *alloc) {
+
+  if (!EliminateAllocations || !alloc->_is_scalar_replaceable) {
+    return false;
+  }
+
+  extract_call_projections(alloc);
+
+  GrowableArray <SafePointNode *> safepoints;
+  if (!can_eliminate_allocation(alloc, safepoints)) {
+    return false;
+  }
+
+  if (!scalar_replacement(alloc, safepoints)) {
+    return false;
+  }
+
+  process_users_of_allocation(alloc);
+
+#ifndef PRODUCT
+if (PrintEliminateAllocations) {
+  if (alloc->is_AllocateArray())
+    tty->print_cr("++++ Eliminated: %d AllocateArray", alloc->_idx);
+  else
+    tty->print_cr("++++ Eliminated: %d Allocate", alloc->_idx);
+}
+#endif
+
+  return true;
+}
+
 
 //---------------------------set_eden_pointers-------------------------
 void PhaseMacroExpand::set_eden_pointers(Node* &eden_top_adr, Node* &eden_end_adr) {
@@ -270,6 +901,13 @@
   Node* klass_node        = alloc->in(AllocateNode::KlassNode);
   Node* initial_slow_test = alloc->in(AllocateNode::InitialTest);
 
+  // With escape analysis, the entire memory state was needed to be able to
+  // eliminate the allocation.  Since the allocations cannot be eliminated,
+  // optimize it to the raw slice.
+  if (mem->is_MergeMem()) {
+    mem = mem->as_MergeMem()->memory_at(Compile::AliasIdxRaw);
+  }
+
   Node* eden_top_adr;
   Node* eden_end_adr;
   set_eden_pointers(eden_top_adr, eden_end_adr);
@@ -813,27 +1451,87 @@
 // Note:  The membar's associated with the lock/unlock are currently not
 //        eliminated.  This should be investigated as a future enhancement.
 //
-void PhaseMacroExpand::eliminate_locking_node(AbstractLockNode *alock) {
-  Node* mem = alock->in(TypeFunc::Memory);
+bool PhaseMacroExpand::eliminate_locking_node(AbstractLockNode *alock) {
+
+  if (!alock->is_eliminated()) {
+    return false;
+  }
+  // Mark the box lock as eliminated if all correspondent locks are eliminated
+  // to construct correct debug info.
+  BoxLockNode* box = alock->box_node()->as_BoxLock();
+  if (!box->is_eliminated()) {
+    bool eliminate = true;
+    for (DUIterator_Fast imax, i = box->fast_outs(imax); i < imax; i++) {
+      Node *lck = box->fast_out(i);
+      if (lck->is_Lock() && !lck->as_AbstractLock()->is_eliminated()) {
+        eliminate = false;
+        break;
+      }
+    }
+    if (eliminate)
+      box->set_eliminated();
+  }
+
+  #ifndef PRODUCT
+  if (PrintEliminateLocks) {
+    if (alock->is_Lock()) {
+      tty->print_cr("++++ Eliminating: %d Lock", alock->_idx);
+    } else {
+      tty->print_cr("++++ Eliminating: %d Unlock", alock->_idx);
+    }
+  }
+  #endif
+
+  Node* mem  = alock->in(TypeFunc::Memory);
+  Node* ctrl = alock->in(TypeFunc::Control);
+
+  extract_call_projections(alock);
+  // There are 2 projections from the lock.  The lock node will
+  // be deleted when its last use is subsumed below.
+  assert(alock->outcnt() == 2 &&
+         _fallthroughproj != NULL &&
+         _memproj_fallthrough != NULL,
+         "Unexpected projections from Lock/Unlock");
+
+  Node* fallthroughproj = _fallthroughproj;
+  Node* memproj_fallthrough = _memproj_fallthrough;
 
   // The memory projection from a lock/unlock is RawMem
   // The input to a Lock is merged memory, so extract its RawMem input
   // (unless the MergeMem has been optimized away.)
   if (alock->is_Lock()) {
-    if (mem->is_MergeMem())
-      mem = mem->as_MergeMem()->in(Compile::AliasIdxRaw);
+    // Seach for MemBarAcquire node and delete it also.
+    MemBarNode* membar = fallthroughproj->unique_ctrl_out()->as_MemBar();
+    assert(membar != NULL && membar->Opcode() == Op_MemBarAcquire, "");
+    Node* ctrlproj = membar->proj_out(TypeFunc::Control);
+    Node* memproj = membar->proj_out(TypeFunc::Memory);
+    _igvn.hash_delete(ctrlproj);
+    _igvn.subsume_node(ctrlproj, fallthroughproj);
+    _igvn.hash_delete(memproj);
+    _igvn.subsume_node(memproj, memproj_fallthrough);
   }
 
-  extract_call_projections(alock);
-  // There are 2 projections from the lock.  The lock node will
-  // be deleted when its last use is subsumed below.
-  assert(alock->outcnt() == 2 && _fallthroughproj != NULL &&
-          _memproj_fallthrough != NULL, "Unexpected projections from Lock/Unlock");
-  _igvn.hash_delete(_fallthroughproj);
-  _igvn.subsume_node(_fallthroughproj, alock->in(TypeFunc::Control));
-  _igvn.hash_delete(_memproj_fallthrough);
-  _igvn.subsume_node(_memproj_fallthrough, mem);
-  return;
+  // Seach for MemBarRelease node and delete it also.
+  if (alock->is_Unlock() && ctrl != NULL && ctrl->is_Proj() &&
+      ctrl->in(0)->is_MemBar()) {
+    MemBarNode* membar = ctrl->in(0)->as_MemBar();
+    assert(membar->Opcode() == Op_MemBarRelease &&
+           mem->is_Proj() && membar == mem->in(0), "");
+    _igvn.hash_delete(fallthroughproj);
+    _igvn.subsume_node(fallthroughproj, ctrl);
+    _igvn.hash_delete(memproj_fallthrough);
+    _igvn.subsume_node(memproj_fallthrough, mem);
+    fallthroughproj = ctrl;
+    memproj_fallthrough = mem;
+    ctrl = membar->in(TypeFunc::Control);
+    mem  = membar->in(TypeFunc::Memory);
+  }
+
+  _igvn.hash_delete(fallthroughproj);
+  _igvn.subsume_node(fallthroughproj, ctrl);
+  _igvn.hash_delete(memproj_fallthrough);
+  _igvn.subsume_node(memproj_fallthrough, mem);
+  return true;
 }
 
 
@@ -844,12 +1542,7 @@
   Node* mem = lock->in(TypeFunc::Memory);
   Node* obj = lock->obj_node();
   Node* box = lock->box_node();
-  Node *flock = lock->fastlock_node();
-
-  if (lock->is_eliminated()) {
-    eliminate_locking_node(lock);
-    return;
-  }
+  Node* flock = lock->fastlock_node();
 
   // Make the merge point
   Node *region = new (C, 3) RegionNode(3);
@@ -898,17 +1591,11 @@
 //------------------------------expand_unlock_node----------------------
 void PhaseMacroExpand::expand_unlock_node(UnlockNode *unlock) {
 
-  Node *ctrl = unlock->in(TypeFunc::Control);
+  Node* ctrl = unlock->in(TypeFunc::Control);
   Node* mem = unlock->in(TypeFunc::Memory);
   Node* obj = unlock->obj_node();
   Node* box = unlock->box_node();
 
-
-  if (unlock->is_eliminated()) {
-    eliminate_locking_node(unlock);
-    return;
-  }
-
   // No need for a null check on unlock
 
   // Make the merge point
@@ -958,14 +1645,41 @@
 bool PhaseMacroExpand::expand_macro_nodes() {
   if (C->macro_count() == 0)
     return false;
-  // Make sure expansion will not cause node limit to be exceeded.  Worst case is a
-  // macro node gets expanded into about 50 nodes.  Allow 50% more for optimization
+  // attempt to eliminate allocations
+  bool progress = true;
+  while (progress) {
+    progress = false;
+    for (int i = C->macro_count(); i > 0; i--) {
+      Node * n = C->macro_node(i-1);
+      bool success = false;
+      debug_only(int old_macro_count = C->macro_count(););
+      switch (n->class_id()) {
+      case Node::Class_Allocate:
+      case Node::Class_AllocateArray:
+        success = eliminate_allocate_node(n->as_Allocate());
+        break;
+      case Node::Class_Lock:
+      case Node::Class_Unlock:
+        success = eliminate_locking_node(n->as_AbstractLock());
+        break;
+      default:
+        assert(false, "unknown node type in macro list");
+      }
+      assert(success == (C->macro_count() < old_macro_count), "elimination reduces macro count");
+      progress = progress || success;
+    }
+  }
+  // Make sure expansion will not cause node limit to be exceeded.
+  // Worst case is a macro node gets expanded into about 50 nodes.
+  // Allow 50% more for optimization.
   if (C->check_node_count(C->macro_count() * 75, "out of nodes before macro expansion" ) )
     return true;
+
   // expand "macro" nodes
   // nodes are removed from the macro list as they are processed
   while (C->macro_count() > 0) {
-    Node * n = C->macro_node(0);
+    int macro_count = C->macro_count();
+    Node * n = C->macro_node(macro_count-1);
     assert(n->is_macro(), "only macro nodes expected here");
     if (_igvn.type(n) == Type::TOP || n->in(0)->is_top() ) {
       // node is unreachable, so don't try to expand it
@@ -988,6 +1702,7 @@
     default:
       assert(false, "unknown node type in macro list");
     }
+    assert(C->macro_count() < macro_count, "must have deleted a node from macro list");
     if (C->failing())  return true;
   }
   _igvn.optimize();
--- a/src/share/vm/opto/macro.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/macro.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -78,7 +78,16 @@
                               Node* length,
                               const TypeFunc* slow_call_type,
                               address slow_call_address);
-  void eliminate_locking_node(AbstractLockNode *alock);
+  Node *value_from_mem(Node *mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc);
+  Node *value_from_mem_phi(Node *mem, BasicType ft, const Type *ftype, const TypeOopPtr *adr_t, Node *alloc, int level);
+
+  bool eliminate_allocate_node(AllocateNode *alloc);
+  bool can_eliminate_allocation(AllocateNode *alloc, GrowableArray <SafePointNode *>& safepoints);
+  bool scalar_replacement(AllocateNode *alloc, GrowableArray <SafePointNode *>& safepoints_done);
+  void process_users_of_allocation(AllocateNode *alloc);
+
+  void eliminate_card_mark(Node *cm);
+  bool eliminate_locking_node(AbstractLockNode *alock);
   void expand_lock_node(LockNode *lock);
   void expand_unlock_node(UnlockNode *unlock);
 
--- a/src/share/vm/opto/matcher.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/matcher.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -1647,6 +1647,7 @@
       case Op_Phi:             // Treat Phis as shared roots
       case Op_Parm:
       case Op_Proj:            // All handled specially during matching
+      case Op_SafePointScalarObject:
         set_shared(n);
         set_dontcare(n);
         break;
--- a/src/share/vm/opto/memnode.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/memnode.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -29,6 +29,8 @@
 #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); }
 
@@ -87,6 +89,112 @@
 
 #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.
@@ -119,48 +227,8 @@
   if (mem->is_MergeMem()) {
     MergeMemNode* mmem = mem->as_MergeMem();
     const TypePtr *tp = t_adr->is_ptr();
-    uint alias_idx = phase->C->get_alias_index(tp);
-#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");
-      const TypePtr *adr_t =  adr_type();
-      bool consistent =  adr_t == NULL || adr_t->empty() || phase->C->must_alias(adr_t, alias_idx );
-      // Sometimes dead array references collapse to a[-1], a[-2], or a[-3]
-      if( !consistent && adr_t != NULL && !adr_t->empty() &&
-             tp->isa_aryptr() &&    tp->offset() == Type::OffsetBot &&
-          adr_t->isa_aryptr() && adr_t->offset() != Type::OffsetBot &&
-          ( adr_t->offset() == arrayOopDesc::length_offset_in_bytes() ||
-            adr_t->offset() == oopDesc::klass_offset_in_bytes() ||
-            adr_t->offset() == oopDesc::mark_offset_in_bytes() ) ) {
-        // don't assert if it is dead code.
-        consistent = true;
-      }
-      if( !consistent ) {
-        tty->print("alias_idx==%d, adr_type()==", alias_idx); if( adr_t == NULL ) { tty->print("NULL"); } else { adr_t->dump(); }
-        tty->cr();
-        print_alias_types();
-        assert(consistent, "adr_type 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 TypeInstPtr *tinst = tp->isa_instptr();
-    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
-    }
+
+    mem = step_through_mergemem(phase, mmem, tp, adr_type(), tty);
   }
 
   if (mem != old_mem) {
@@ -254,6 +322,8 @@
   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...
@@ -333,6 +403,22 @@
         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,
@@ -534,7 +620,10 @@
           const Node* call = adr->in(0);
           if (call->is_CallStaticJava()) {
             const CallStaticJavaNode* call_java = call->as_CallStaticJava();
-            assert(call_java && call_java->method() == NULL, "must be runtime call");
+            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.
@@ -732,6 +821,21 @@
   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 ) {
@@ -754,6 +858,25 @@
     // 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;
 }
 
@@ -962,6 +1085,122 @@
     }
   }
 
+  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");
+        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
@@ -1189,6 +1428,17 @@
       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;
 }
 
@@ -1712,7 +1962,7 @@
   const TypeOopPtr *adr_oop = phase->type(adr)->isa_oopptr();
   if (adr_oop == NULL)
     return false;
-  if (!adr_oop->is_instance())
+  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);
@@ -1821,7 +2071,7 @@
 //------------------------------Identity---------------------------------------
 // Clearing a zero length array does nothing
 Node *ClearArrayNode::Identity( PhaseTransform *phase ) {
-  return phase->type(in(2))->higher_equal(TypeInt::ZERO)  ? in(1) : this;
+  return phase->type(in(2))->higher_equal(TypeX::ZERO)  ? in(1) : this;
 }
 
 //------------------------------Idealize---------------------------------------
@@ -1894,6 +2144,11 @@
                                    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;
@@ -1919,6 +2174,11 @@
                                    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;
@@ -3244,7 +3504,7 @@
     }
   }
 
-  assert(verify_sparse(), "please, no dups of base");
+  assert(progress || verify_sparse(), "please, no dups of base");
   return progress;
 }
 
--- a/src/share/vm/opto/memnode.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/memnode.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -67,6 +67,8 @@
                                       PhaseTransform* phase);
   static bool adr_phi_is_loop_invariant(Node* adr_phi, Node* cast);
 
+  static Node *optimize_simple_memory_chain(Node *mchain, const TypePtr *t_adr, PhaseGVN *phase);
+  static Node *optimize_memory_chain(Node *mchain, const TypePtr *t_adr, PhaseGVN *phase);
   // This one should probably be a phase-specific function:
   static bool detect_dominating_control(Node* dom, Node* sub);
 
@@ -172,6 +174,9 @@
   // Map a load opcode to its corresponding store opcode.
   virtual int store_Opcode() const = 0;
 
+  // Check if the load's memory input is a Phi node with the same control.
+  bool is_instance_field_load_with_local_phi(Node* ctrl);
+
 #ifndef PRODUCT
   virtual void dump_spec(outputStream *st) const;
 #endif
--- a/src/share/vm/opto/node.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/node.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -812,8 +812,7 @@
 Node* Node::uncast() const {
   // Should be inline:
   //return is_ConstraintCast() ? uncast_helper(this) : (Node*) this;
-  if (is_ConstraintCast() ||
-      (is_Type() && req() == 2 && Opcode() == Op_CheckCastPP))
+  if (is_ConstraintCast() || is_CheckCastPP())
     return uncast_helper(this);
   else
     return (Node*) this;
@@ -827,7 +826,7 @@
       break;
     } else if (p->is_ConstraintCast()) {
       p = p->in(1);
-    } else if (p->is_Type() && p->Opcode() == Op_CheckCastPP) {
+    } else if (p->is_CheckCastPP()) {
       p = p->in(1);
     } else {
       break;
--- a/src/share/vm/opto/node.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/node.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -106,6 +106,7 @@
 class RegionNode;
 class RootNode;
 class SafePointNode;
+class SafePointScalarObjectNode;
 class StartNode;
 class State;
 class StoreNode;
@@ -575,6 +576,7 @@
       DEFINE_CLASS_ID(ConstraintCast, Type, 1)
       DEFINE_CLASS_ID(CheckCastPP, Type, 2)
       DEFINE_CLASS_ID(CMove, Type, 3)
+      DEFINE_CLASS_ID(SafePointScalarObject, Type, 4)
 
     DEFINE_CLASS_ID(Mem,   Node, 6)
       DEFINE_CLASS_ID(Load,  Mem, 0)
@@ -721,6 +723,7 @@
   DEFINE_CLASS_QUERY(Region)
   DEFINE_CLASS_QUERY(Root)
   DEFINE_CLASS_QUERY(SafePoint)
+  DEFINE_CLASS_QUERY(SafePointScalarObject)
   DEFINE_CLASS_QUERY(Start)
   DEFINE_CLASS_QUERY(Store)
   DEFINE_CLASS_QUERY(Sub)
@@ -1325,7 +1328,6 @@
 // Inline definition of Compile::record_for_igvn must be deferred to this point.
 inline void Compile::record_for_igvn(Node* n) {
   _for_igvn->push(n);
-  record_for_escape_analysis(n);
 }
 
 //------------------------------Node_Stack-------------------------------------
--- a/src/share/vm/opto/output.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/output.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -561,7 +561,30 @@
     : new LocationValue(Location::new_stk_loc(l_type,  ra->reg2offset(regnum)));
 }
 
-void Compile::FillLocArray( int idx, Node *local, GrowableArray<ScopeValue*> *array ) {
+
+ObjectValue*
+Compile::sv_for_node_id(GrowableArray<ScopeValue*> *objs, int id) {
+  for (int i = 0; i < objs->length(); i++) {
+    assert(objs->at(i)->is_object(), "corrupt object cache");
+    ObjectValue* sv = (ObjectValue*) objs->at(i);
+    if (sv->id() == id) {
+      return sv;
+    }
+  }
+  // Otherwise..
+  return NULL;
+}
+
+void Compile::set_sv_for_object_node(GrowableArray<ScopeValue*> *objs,
+                                     ObjectValue* sv ) {
+  assert(sv_for_node_id(objs, sv->id()) == NULL, "Precondition");
+  objs->append(sv);
+}
+
+
+void Compile::FillLocArray( int idx, MachSafePointNode* sfpt, Node *local,
+                            GrowableArray<ScopeValue*> *array,
+                            GrowableArray<ScopeValue*> *objs ) {
   assert( local, "use _top instead of null" );
   if (array->length() != idx) {
     assert(array->length() == idx + 1, "Unexpected array count");
@@ -578,6 +601,29 @@
   }
   const Type *t = local->bottom_type();
 
+  // Is it a safepoint scalar object node?
+  if (local->is_SafePointScalarObject()) {
+    SafePointScalarObjectNode* spobj = local->as_SafePointScalarObject();
+
+    ObjectValue* sv = Compile::sv_for_node_id(objs, spobj->_idx);
+    if (sv == NULL) {
+      ciKlass* cik = t->is_oopptr()->klass();
+      assert(cik->is_instance_klass() ||
+             cik->is_array_klass(), "Not supported allocation.");
+      sv = new ObjectValue(spobj->_idx,
+                           new ConstantOopWriteValue(cik->encoding()));
+      Compile::set_sv_for_object_node(objs, sv);
+
+      uint first_ind = spobj->first_index();
+      for (uint i = 0; i < spobj->n_fields(); i++) {
+        Node* fld_node = sfpt->in(first_ind+i);
+        (void)FillLocArray(sv->field_values()->length(), sfpt, fld_node, sv->field_values(), objs);
+      }
+    }
+    array->append(sv);
+    return;
+  }
+
   // Grab the register number for the local
   OptoReg::Name regnum = _regalloc->get_reg_first(local);
   if( OptoReg::is_valid(regnum) ) {// Got a register/stack?
@@ -755,6 +801,11 @@
   JVMState* youngest_jvms = sfn->jvms();
   int max_depth = youngest_jvms->depth();
 
+  // Allocate the object pool for scalar-replaced objects -- the map from
+  // small-integer keys (which can be recorded in the local and ostack
+  // arrays) to descriptions of the object state.
+  GrowableArray<ScopeValue*> *objs = new GrowableArray<ScopeValue*>();
+
   // Visit scopes from oldest to youngest.
   for (int depth = 1; depth <= max_depth; depth++) {
     JVMState* jvms = youngest_jvms->of_depth(depth);
@@ -773,13 +824,13 @@
     // Insert locals into the locarray
     GrowableArray<ScopeValue*> *locarray = new GrowableArray<ScopeValue*>(num_locs);
     for( idx = 0; idx < num_locs; idx++ ) {
-      FillLocArray( idx, sfn->local(jvms, idx), locarray );
+      FillLocArray( idx, sfn, sfn->local(jvms, idx), locarray, objs );
     }
 
     // Insert expression stack entries into the exparray
     GrowableArray<ScopeValue*> *exparray = new GrowableArray<ScopeValue*>(num_exps);
     for( idx = 0; idx < num_exps; idx++ ) {
-      FillLocArray( idx,  sfn->stack(jvms, idx), exparray );
+      FillLocArray( idx,  sfn, sfn->stack(jvms, idx), exparray, objs );
     }
 
     // Add in mappings of the monitors
@@ -803,7 +854,27 @@
 
       // Create ScopeValue for object
       ScopeValue *scval = NULL;
-      if( !obj_node->is_Con() ) {
+
+      if( obj_node->is_SafePointScalarObject() ) {
+        SafePointScalarObjectNode* spobj = obj_node->as_SafePointScalarObject();
+        scval = Compile::sv_for_node_id(objs, spobj->_idx);
+        if (scval == NULL) {
+          const Type *t = obj_node->bottom_type();
+          ciKlass* cik = t->is_oopptr()->klass();
+          assert(cik->is_instance_klass() ||
+                 cik->is_array_klass(), "Not supported allocation.");
+          ObjectValue* sv = new ObjectValue(spobj->_idx,
+                                new ConstantOopWriteValue(cik->encoding()));
+          Compile::set_sv_for_object_node(objs, sv);
+
+          uint first_ind = spobj->first_index();
+          for (uint i = 0; i < spobj->n_fields(); i++) {
+            Node* fld_node = sfn->in(first_ind+i);
+            (void)FillLocArray(sv->field_values()->length(), sfn, fld_node, sv->field_values(), objs);
+          }
+          scval = sv;
+        }
+      } else if( !obj_node->is_Con() ) {
         OptoReg::Name obj_reg = _regalloc->get_reg_first(obj_node);
         scval = new_loc_value( _regalloc, obj_reg, Location::oop );
       } else {
@@ -811,9 +882,13 @@
       }
 
       OptoReg::Name box_reg = BoxLockNode::stack_slot(box_node);
-      monarray->append(new MonitorValue(scval, Location::new_stk_loc(Location::normal,_regalloc->reg2offset(box_reg))));
+      Location basic_lock = Location::new_stk_loc(Location::normal,_regalloc->reg2offset(box_reg));
+      monarray->append(new MonitorValue(scval, basic_lock, box_node->as_BoxLock()->is_eliminated()));
     }
 
+    // We dump the object pool first, since deoptimization reads it in first.
+    debug_info()->dump_object_pool(objs);
+
     // Build first class objects to pass to scope
     DebugToken *locvals = debug_info()->create_scope_values(locarray);
     DebugToken *expvals = debug_info()->create_scope_values(exparray);
@@ -823,6 +898,7 @@
     ciMethod* scope_method = method ? method : _method;
     // Describe the scope here
     assert(jvms->bci() >= InvocationEntryBci && jvms->bci() <= 0x10000, "must be a valid or entry BCI");
+    // Now we can describe the scope.
     debug_info()->describe_scope(safepoint_pc_offset,scope_method,jvms->bci(),locvals,expvals,monvals);
   } // End jvms loop
 
--- a/src/share/vm/opto/phaseX.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/phaseX.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -587,11 +587,6 @@
       Node_Notes* loc = C->locate_node_notes(nna, x->_idx, true);
       loc->clear(); // do not put debug info on constants
     }
-    // Collect points-to information for escape analysys
-    ConnectionGraph *cgr = C->congraph();
-    if (cgr != NULL) {
-      cgr->record_escape(x, this);
-    }
   } else {
     x->destruct();              // Hit, destroy duplicate constant
     x = k;                      // use existing constant
@@ -714,12 +709,6 @@
     return i;
   }
 
-  // Collect points-to information for escape analysys
-  ConnectionGraph *cgr = C->congraph();
-  if (cgr != NULL) {
-    cgr->record_escape(k, this);
-  }
-
   // Return Idealized original
   return k;
 }
@@ -1245,7 +1234,7 @@
 
     uint use_op = use->Opcode();
     // If changed Cast input, check Phi users for simple cycles
-    if( use->is_ConstraintCast() || use->Opcode() == Op_CheckCastPP ) {
+    if( use->is_ConstraintCast() || use->is_CheckCastPP() ) {
       for (DUIterator_Fast i2max, i2 = use->fast_outs(i2max); i2 < i2max; i2++) {
         Node* u = use->fast_out(i2);
         if (u->is_Phi())
--- a/src/share/vm/opto/phaseX.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/phaseX.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -439,6 +439,13 @@
   void add_users_to_worklist0( Node *n );
   void add_users_to_worklist ( Node *n );
 
+  // Replace old node with new one.
+  void replace_node( Node *old, Node *nn ) {
+    add_users_to_worklist(old);
+    hash_delete(old);
+    subsume_node(old, nn);
+  }
+
 #ifndef PRODUCT
 protected:
   // Sub-quadratic implementation of VerifyIterativeGVN.
--- a/src/share/vm/opto/postaloc.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/postaloc.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -253,7 +253,8 @@
 // nodes can represent the same constant so the type and rule of the
 // MachNode must be checked to ensure equivalence.
 //
-bool PhaseChaitin::eliminate_copy_of_constant(Node* val, Block *current_block,
+bool PhaseChaitin::eliminate_copy_of_constant(Node* val, Node* n,
+                                              Block *current_block,
                                               Node_List& value, Node_List& regnd,
                                               OptoReg::Name nreg, OptoReg::Name nreg2) {
   if (value[nreg] != val && val->is_Con() &&
@@ -269,12 +270,12 @@
     // Since they are equivalent the second one if redundant and can
     // be removed.
     //
-    // val will be replaced with the old value but val might have
+    // n will be replaced with the old value but n might have
     // kills projections associated with it so remove them now so that
     // yank_if_dead will be able to elminate the copy once the uses
     // have been transferred to the old[value].
-    for (DUIterator_Fast imax, i = val->fast_outs(imax); i < imax; i++) {
-      Node* use = val->fast_out(i);
+    for (DUIterator_Fast imax, i = n->fast_outs(imax); i < imax; i++) {
+      Node* use = n->fast_out(i);
       if (use->is_Proj() && use->outcnt() == 0) {
         // Kill projections have no users and one input
         use->set_req(0, C->top());
@@ -521,7 +522,7 @@
         // then 'n' is a useless copy.  Do not update the register->node
         // mapping so 'n' will go dead.
         if( value[nreg] != val ) {
-          if (eliminate_copy_of_constant(val, b, value, regnd, nreg, OptoReg::Bad)) {
+          if (eliminate_copy_of_constant(val, n, b, value, regnd, nreg, OptoReg::Bad)) {
             n->replace_by(regnd[nreg]);
             j -= yank_if_dead(n,b,&value,&regnd);
           } else {
@@ -549,7 +550,7 @@
           nreg_lo = tmp.find_first_elem();
         }
         if( value[nreg] != val || value[nreg_lo] != val ) {
-          if (eliminate_copy_of_constant(n, b, value, regnd, nreg, nreg_lo)) {
+          if (eliminate_copy_of_constant(val, n, b, value, regnd, nreg, nreg_lo)) {
             n->replace_by(regnd[nreg]);
             j -= yank_if_dead(n,b,&value,&regnd);
           } else {
--- a/src/share/vm/opto/superword.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/superword.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -183,8 +183,8 @@
 
 #ifndef PRODUCT
   if (TraceSuperWord)
-    tty->print_cr("\noffset = %d iv_adjustment = %d  elt_align = %d",
-                  offset, iv_adjustment, align_to_ref_p.memory_size());
+    tty->print_cr("\noffset = %d iv_adjustment = %d  elt_align = %d scale = %d iv_stride = %d",
+                  offset, iv_adjustment, align_to_ref_p.memory_size(), align_to_ref_p.scale_in_bytes(), iv_stride());
 #endif
 
   // Set alignment relative to "align_to_ref"
@@ -1543,7 +1543,7 @@
   Node *pre_opaq1 = pre_end->limit();
   assert(pre_opaq1->Opcode() == Op_Opaque1, "");
   Opaque1Node *pre_opaq = (Opaque1Node*)pre_opaq1;
-  Node *pre_limit = pre_opaq->in(1);
+  Node *lim0 = pre_opaq->in(1);
 
   // Where we put new limit calculations
   Node *pre_ctrl = pre_end->loopnode()->in(LoopNode::EntryControl);
@@ -1555,64 +1555,116 @@
 
   SWPointer align_to_ref_p(align_to_ref, this);
 
-  // Let l0 == original pre_limit, l == new pre_limit, V == v_align
+  // Given:
+  //     lim0 == original pre loop limit
+  //     V == v_align (power of 2)
+  //     invar == extra invariant piece of the address expression
+  //     e == k [ +/- invar ]
+  //
+  // When reassociating expressions involving '%' the basic rules are:
+  //     (a - b) % k == 0   =>  a % k == b % k
+  // and:
+  //     (a + b) % k == 0   =>  a % k == (k - b) % k
+  //
+  // For stride > 0 && scale > 0,
+  //   Derive the new pre-loop limit "lim" such that the two constraints:
+  //     (1) lim = lim0 + N           (where N is some positive integer < V)
+  //     (2) (e + lim) % V == 0
+  //   are true.
+  //
+  //   Substituting (1) into (2),
+  //     (e + lim0 + N) % V == 0
+  //   solve for N:
+  //     N = (V - (e + lim0)) % V
+  //   substitute back into (1), so that new limit
+  //     lim = lim0 + (V - (e + lim0)) % V
   //
-  // For stride > 0
-  //   Need l such that l > l0 && (l+k)%V == 0
-  //   Find n such that l = (l0 + n)
-  //   (l0 + n + k) % V == 0
-  //   n = [V - (l0 + k)%V]%V
-  //   new limit = l0 + [V - (l0 + k)%V]%V
-  // For stride < 0
-  //   Need l such that l < l0 && (l+k)%V == 0
-  //   Find n such that l = (l0 - n)
-  //   (l0 - n + k) % V == 0
-  //   n = (l0 + k)%V
-  //   new limit = l0 - (l0 + k)%V
+  // For stride > 0 && scale < 0
+  //   Constraints:
+  //     lim = lim0 + N
+  //     (e - lim) % V == 0
+  //   Solving for lim:
+  //     (e - lim0 - N) % V == 0
+  //     N = (e - lim0) % V
+  //     lim = lim0 + (e - lim0) % V
+  //
+  // For stride < 0 && scale > 0
+  //   Constraints:
+  //     lim = lim0 - N
+  //     (e + lim) % V == 0
+  //   Solving for lim:
+  //     (e + lim0 - N) % V == 0
+  //     N = (e + lim0) % V
+  //     lim = lim0 - (e + lim0) % V
+  //
+  // For stride < 0 && scale < 0
+  //   Constraints:
+  //     lim = lim0 - N
+  //     (e - lim) % V == 0
+  //   Solving for lim:
+  //     (e - lim0 + N) % V == 0
+  //     N = (V - (e - lim0)) % V
+  //     lim = lim0 - (V - (e - lim0)) % V
 
+  int stride   = iv_stride();
+  int scale    = align_to_ref_p.scale_in_bytes();
   int elt_size = align_to_ref_p.memory_size();
   int v_align  = vector_width_in_bytes() / elt_size;
   int k        = align_to_ref_p.offset_in_bytes() / elt_size;
 
   Node *kn   = _igvn.intcon(k);
-  Node *limk = new (_phase->C, 3) AddINode(pre_limit, kn);
-  _phase->_igvn.register_new_node_with_optimizer(limk);
-  _phase->set_ctrl(limk, pre_ctrl);
+
+  Node *e = kn;
   if (align_to_ref_p.invar() != NULL) {
+    // incorporate any extra invariant piece producing k +/- invar >>> log2(elt)
     Node* log2_elt = _igvn.intcon(exact_log2(elt_size));
     Node* aref     = new (_phase->C, 3) URShiftINode(align_to_ref_p.invar(), log2_elt);
     _phase->_igvn.register_new_node_with_optimizer(aref);
     _phase->set_ctrl(aref, pre_ctrl);
-    if (!align_to_ref_p.negate_invar()) {
-      limk = new (_phase->C, 3) AddINode(limk, aref);
+    if (align_to_ref_p.negate_invar()) {
+      e = new (_phase->C, 3) SubINode(e, aref);
     } else {
-      limk = new (_phase->C, 3) SubINode(limk, aref);
+      e = new (_phase->C, 3) AddINode(e, aref);
     }
-    _phase->_igvn.register_new_node_with_optimizer(limk);
-    _phase->set_ctrl(limk, pre_ctrl);
+    _phase->_igvn.register_new_node_with_optimizer(e);
+    _phase->set_ctrl(e, pre_ctrl);
   }
-  Node* va_msk = _igvn.intcon(v_align - 1);
-  Node* n      = new (_phase->C, 3) AndINode(limk, va_msk);
-  _phase->_igvn.register_new_node_with_optimizer(n);
-  _phase->set_ctrl(n, pre_ctrl);
-  Node* newlim;
-  if (iv_stride() > 0) {
+
+  // compute e +/- lim0
+  if (scale < 0) {
+    e = new (_phase->C, 3) SubINode(e, lim0);
+  } else {
+    e = new (_phase->C, 3) AddINode(e, lim0);
+  }
+  _phase->_igvn.register_new_node_with_optimizer(e);
+  _phase->set_ctrl(e, pre_ctrl);
+
+  if (stride * scale > 0) {
+    // compute V - (e +/- lim0)
     Node* va  = _igvn.intcon(v_align);
-    Node* adj = new (_phase->C, 3) SubINode(va, n);
-    _phase->_igvn.register_new_node_with_optimizer(adj);
-    _phase->set_ctrl(adj, pre_ctrl);
-    Node* adj2 = new (_phase->C, 3) AndINode(adj, va_msk);
-    _phase->_igvn.register_new_node_with_optimizer(adj2);
-    _phase->set_ctrl(adj2, pre_ctrl);
-    newlim = new (_phase->C, 3) AddINode(pre_limit, adj2);
+    e = new (_phase->C, 3) SubINode(va, e);
+    _phase->_igvn.register_new_node_with_optimizer(e);
+    _phase->set_ctrl(e, pre_ctrl);
+  }
+  // compute N = (exp) % V
+  Node* va_msk = _igvn.intcon(v_align - 1);
+  Node* N = new (_phase->C, 3) AndINode(e, va_msk);
+  _phase->_igvn.register_new_node_with_optimizer(N);
+  _phase->set_ctrl(N, pre_ctrl);
+
+  //   substitute back into (1), so that new limit
+  //     lim = lim0 + N
+  Node* lim;
+  if (stride < 0) {
+    lim = new (_phase->C, 3) SubINode(lim0, N);
   } else {
-    newlim = new (_phase->C, 3) SubINode(pre_limit, n);
+    lim = new (_phase->C, 3) AddINode(lim0, N);
   }
-  _phase->_igvn.register_new_node_with_optimizer(newlim);
-  _phase->set_ctrl(newlim, pre_ctrl);
+  _phase->_igvn.register_new_node_with_optimizer(lim);
+  _phase->set_ctrl(lim, pre_ctrl);
   Node* constrained =
-    (iv_stride() > 0) ? (Node*) new (_phase->C,3) MinINode(newlim, orig_limit)
-                      : (Node*) new (_phase->C,3) MaxINode(newlim, orig_limit);
+    (stride > 0) ? (Node*) new (_phase->C,3) MinINode(lim, orig_limit)
+                 : (Node*) new (_phase->C,3) MaxINode(lim, orig_limit);
   _phase->_igvn.register_new_node_with_optimizer(constrained);
   _phase->set_ctrl(constrained, pre_ctrl);
   _igvn.hash_delete(pre_opaq);
--- a/src/share/vm/opto/type.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/type.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -3164,7 +3164,7 @@
     case TopPTR:
       // Compute new klass on demand, do not use tap->_klass
       xk = (tap->_klass_is_exact | this->_klass_is_exact);
-      return make( ptr, const_oop(), tary, lazy_klass, xk, off );
+      return make( ptr, const_oop(), tary, lazy_klass, xk, off, iid );
     case Constant: {
       ciObject* o = const_oop();
       if( _ptr == Constant ) {
@@ -3176,7 +3176,7 @@
         o = tap->const_oop();
       }
       xk = true;
-      return TypeAryPtr::make( ptr, o, tary, tap->_klass, xk, off );
+      return TypeAryPtr::make( ptr, o, tary, tap->_klass, xk, off, iid );
     }
     case NotNull:
     case BotPTR:
@@ -3263,14 +3263,21 @@
     break;
   }
 
-  st->print("*");
+  if( _offset != 0 ) {
+    int header_size = objArrayOopDesc::header_size() * wordSize;
+    if( _offset == OffsetTop )       st->print("+undefined");
+    else if( _offset == OffsetBot )  st->print("+any");
+    else if( _offset < header_size ) st->print("+%d", _offset);
+    else {
+      BasicType basic_elem_type = elem()->basic_type();
+      int array_base = arrayOopDesc::base_offset_in_bytes(basic_elem_type);
+      int elem_size = type2aelembytes(basic_elem_type);
+      st->print("[%d]", (_offset - array_base)/elem_size);
+    }
+  }
+  st->print(" *");
   if (_instance_id != UNKNOWN_INSTANCE)
     st->print(",iid=%d",_instance_id);
-  if( !_offset ) return;
-  if( _offset == OffsetTop )      st->print("+undefined");
-  else if( _offset == OffsetBot ) st->print("+any");
-  else if( _offset < 12 )         st->print("+%d",_offset);
-  else                            st->print("[%d]", (_offset-12)/4 );
 }
 #endif
 
--- a/src/share/vm/opto/type.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/opto/type.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -686,6 +686,7 @@
   bool klass_is_exact()    const { return _klass_is_exact; }
   bool is_instance()       const { return _instance_id != UNKNOWN_INSTANCE; }
   uint instance_id()       const { return _instance_id; }
+  bool is_instance_field() const { return _instance_id != UNKNOWN_INSTANCE && _offset >= 0; }
 
   virtual intptr_t get_con() const;
 
--- a/src/share/vm/runtime/arguments.cpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/runtime/arguments.cpp	Fri Mar 21 08:32:17 2008 -0700
@@ -1276,6 +1276,9 @@
     sprintf(buffer, "java.lang.Integer.IntegerCache.high=%d", AutoBoxCacheMax);
     add_property(buffer);
   }
+  if (AggressiveOpts && FLAG_IS_DEFAULT(DoEscapeAnalysis)) {
+    FLAG_SET_DEFAULT(DoEscapeAnalysis, true);
+  }
 #endif
 
   if (AggressiveOpts) {
--- a/src/share/vm/runtime/globals.hpp	Thu Mar 20 09:17:30 2008 -0500
+++ b/src/share/vm/runtime/globals.hpp	Fri Mar 21 08:32:17 2008 -0700
@@ -943,6 +943,12 @@
   product(bool, UseXmmRegToRegMoveAll, false,                               \
           "Copy all XMM register bits when moving value between registers") \
                                                                             \
+  product(bool, UseXmmI2D, false,                                           \
+          "Use SSE2 CVTDQ2PD instruction to convert Integer to Double")     \
+                                                                            \
+  product(bool, UseXmmI2F, false,                                           \
+          "Use SSE2 CVTDQ2PS instruction to convert Integer to Float")      \
+                                                                            \
   product(intx, FieldsAllocationStyle, 1,                                   \
           "0 - type based with oops first, 1 - with oops last")             \
                                                                             \
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/test/compiler/6659207/Test.java	Fri Mar 21 08:32:17 2008 -0700
@@ -0,0 +1,60 @@
+/*
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * SUN PROPRIETARY/CONFIDENTIAL. Use is subject to license terms.
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ */
+
+/*
+ * @test
+ * @bug 6659207
+ * @summary access violation in CompilerThread0
+ */
+
+public class Test {
+    static int[] array = new int[12];
+
+    static int index(int i) {
+        if (i == 0) return 0;
+        for (int n = 0; n < array.length; n++)
+            if (i < array[n]) return n;
+        return -1;
+    }
+
+    static int test(int i) {
+        int result = 0;
+        i = index(i);
+        if (i >= 0)
+            if (array[i] != 0)
+                result++;
+
+        if (i != -1)
+            array[i]++;
+
+        return result;
+    }
+
+    public static void main(String[] args) {
+        int total = 0;
+        for (int i = 0; i < 100000; i++) {
+            total += test(10);
+        }
+        System.out.println(total);
+    }
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/test/compiler/6661247/Test.java	Fri Mar 21 08:32:17 2008 -0700
@@ -0,0 +1,155 @@
+/*
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * SUN PROPRIETARY/CONFIDENTIAL. Use is subject to license terms.
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ */
+
+/*
+ * @test
+ * @bug 6661247
+ * @summary Internal bug in 32-bit HotSpot optimizer while bit manipulations
+ */
+
+import java.util.Random;
+import java.nio.*;
+
+// This isn't a completely reliable test for 6661247 since the results
+// depend on what the local schedule looks like but it does reproduce
+// the issue in current builds.
+
+public class Test {
+
+    public static void test(boolean[] src, int srcPos, LongBuffer dest, long destPos, int count) {
+        int countStart = (destPos & 63) == 0 ? 0 : 64 - (int)(destPos & 63);
+        if (countStart > count)
+            countStart = count;
+        for (int srcPosMax = srcPos + countStart; srcPos < srcPosMax; srcPos++, destPos++) {
+            if (src[srcPos])
+                dest.put((int)(destPos >>> 6), dest.get((int)(destPos >>> 6)) | 1L << (destPos & 63));
+            else
+                dest.put((int)(destPos >>> 6), dest.get((int)(destPos >>> 6)) & ~(1L << (destPos & 63)));
+        }
+        count -= countStart;
+        int cnt = count >>> 6;
+        for (int k = (int)(destPos >>> 6), kMax = k + cnt; k < kMax; k++) {
+            int low = (src[srcPos] ? 1 : 0)
+                | (src[srcPos + 1] ? 1 << 1 : 0)
+                | (src[srcPos + 2] ? 1 << 2 : 0)
+                | (src[srcPos + 3] ? 1 << 3 : 0)
+                | (src[srcPos + 4] ? 1 << 4 : 0)
+                | (src[srcPos + 5] ? 1 << 5 : 0)
+                | (src[srcPos + 6] ? 1 << 6 : 0)
+                | (src[srcPos + 7] ? 1 << 7 : 0)
+                | (src[srcPos + 8] ? 1 << 8 : 0)
+                | (src[srcPos + 9] ? 1 << 9 : 0)
+                | (src[srcPos + 10] ? 1 << 10 : 0)
+                | (src[srcPos + 11] ? 1 << 11 : 0)
+                | (src[srcPos + 12] ? 1 << 12 : 0)
+                | (src[srcPos + 13] ? 1 << 13 : 0)
+                | (src[srcPos + 14] ? 1 << 14 : 0)
+                | (src[srcPos + 15] ? 1 << 15 : 0)
+                | (src[srcPos + 16] ? 1 << 16 : 0)
+                | (src[srcPos + 17] ? 1 << 17 : 0)
+                | (src[srcPos + 18] ? 1 << 18 : 0)
+                | (src[srcPos + 19] ? 1 << 19 : 0)
+                | (src[srcPos + 20] ? 1 << 20 : 0)
+                | (src[srcPos + 21] ? 1 << 21 : 0)
+                | (src[srcPos + 22] ? 1 << 22 : 0)
+                | (src[srcPos + 23] ? 1 << 23 : 0)
+                | (src[srcPos + 24] ? 1 << 24 : 0)
+                | (src[srcPos + 25] ? 1 << 25 : 0)
+                | (src[srcPos + 26] ? 1 << 26 : 0)
+                | (src[srcPos + 27] ? 1 << 27 : 0)
+                | (src[srcPos + 28] ? 1 << 28 : 0)
+                | (src[srcPos + 29] ? 1 << 29 : 0)
+                | (src[srcPos + 30] ? 1 << 30 : 0)
+                | (src[srcPos + 31] ? 1 << 31 : 0)
+                ;
+            srcPos += 32;
+            int high = (src[srcPos] ? 1 : 0)        // PROBLEM!
+                | (src[srcPos + 1] ? 1 << 1 : 0)
+                | (src[srcPos + 2] ? 1 << 2 : 0)
+                | (src[srcPos + 3] ? 1 << 3 : 0)
+                | (src[srcPos + 4] ? 1 << 4 : 0)
+                | (src[srcPos + 5] ? 1 << 5 : 0)
+                | (src[srcPos + 6] ? 1 << 6 : 0)
+                | (src[srcPos + 7] ? 1 << 7 : 0)
+                | (src[srcPos + 8] ? 1 << 8 : 0)
+                | (src[srcPos + 9] ? 1 << 9 : 0)
+                | (src[srcPos + 10] ? 1 << 10 : 0)
+                | (src[srcPos + 11] ? 1 << 11 : 0)
+                | (src[srcPos + 12] ? 1 << 12 : 0)
+                | (src[srcPos + 13] ? 1 << 13 : 0)
+                | (src[srcPos + 14] ? 1 << 14 : 0)
+                | (src[srcPos + 15] ? 1 << 15 : 0)
+                | (src[srcPos + 16] ? 1 << 16 : 0)
+                | (src[srcPos + 17] ? 1 << 17 : 0)
+                | (src[srcPos + 18] ? 1 << 18 : 0)
+                | (src[srcPos + 19] ? 1 << 19 : 0)
+                | (src[srcPos + 20] ? 1 << 20 : 0)
+                | (src[srcPos + 21] ? 1 << 21 : 0)
+                | (src[srcPos + 22] ? 1 << 22 : 0)
+                | (src[srcPos + 23] ? 1 << 23 : 0)
+                | (src[srcPos + 24] ? 1 << 24 : 0)
+                | (src[srcPos + 25] ? 1 << 25 : 0)
+                | (src[srcPos + 26] ? 1 << 26 : 0)
+                | (src[srcPos + 27] ? 1 << 27 : 0)
+                | (src[srcPos + 28] ? 1 << 28 : 0)
+                | (src[srcPos + 29] ? 1 << 29 : 0)
+                | (src[srcPos + 30] ? 1 << 30 : 0)
+                | (src[srcPos + 31] ? 1 << 31 : 0)
+                ;
+            srcPos += 32;
+            dest.put(k, ((long)low & 0xFFFFFFFFL) | (((long)high) << 32));
+            destPos += 64;
+        }
+        int countFinish = count & 63;
+        for (int srcPosMax = srcPos + countFinish; srcPos < srcPosMax; srcPos++, destPos++) {
+            if (src[srcPos])
+                dest.put((int)(destPos >>> 6), dest.get((int)(destPos >>> 6)) | 1L << (destPos & 63));
+            else
+                dest.put((int)(destPos >>> 6), dest.get((int)(destPos >>> 6)) & ~(1L << (destPos & 63)));
+        }
+    }
+    public static void main(String[] args) {
+        Random r = new Random();
+        int entries = 1000;
+        boolean[] src = new boolean[entries * 64];
+        long[] dest = new long[entries];
+        long[] result = new long[entries];
+
+        for (int c = 0; c < 2000; c++) {
+            for (int i = 0; i < entries; i++) {
+                long l = r.nextLong();
+                for (int bit = 0; bit < 64; bit++) {
+                    src[i * 64 + bit] = (l & (1L << bit)) != 0;
+                }
+                dest[i] = 0;
+                result[i] = l;
+            }
+            test(src, 0, LongBuffer.wrap(dest, 0, dest.length), 0, src.length);
+            for (int i = 0; i < entries; i++) {
+                if (dest[i] != result[i]) {
+                    throw new InternalError(i + ": " + Long.toHexString(dest[i]) + " != " + Long.toHexString(result[i]));
+                }
+            }
+        }
+    }
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/test/compiler/6663621/IVTest.java	Fri Mar 21 08:32:17 2008 -0700
@@ -0,0 +1,116 @@
+/*
+ * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
+ * SUN PROPRIETARY/CONFIDENTIAL. Use is subject to license terms.
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ *
+ */
+
+/**
+ * @test
+ * @bug 6663621
+ * @summary JVM crashes while trying to execute api/java_security/Signature/SignatureTests.html#initSign tests.
+ */
+
+public class IVTest {
+    static int paddedSize;
+
+    static void padV15(byte[] padded) {
+        int psSize = padded.length;
+        int k = 0;
+        while (psSize-- > 0) {
+            padded[k++] = (byte)0xff;
+        }
+    }
+
+    static void padV15_2(int paddedSize) {
+        byte[] padded = new byte[paddedSize];
+        int psSize = padded.length;
+        int k = 0;
+        while (psSize-- > 0) {
+            padded[k++] = (byte)0xff;
+        }
+    }
+
+    static void padV15_3() {
+        byte[] padded = new byte[paddedSize];
+        int psSize = padded.length;
+        int k = 0;
+        while (psSize-- > 0) {
+            padded[k++] = (byte)0xff;
+        }
+    }
+
+    static void padV15_4() {
+        byte[] padded = new byte[paddedSize];
+        int psSize = padded.length;
+        for (int k = 0;psSize > 0; psSize--) {
+            int i = padded.length - psSize;
+            padded[i] = (byte)0xff;
+        }
+    }
+
+    static void padV15_5() {
+        byte[] padded = new byte[paddedSize];
+        int psSize = padded.length;
+        int k = psSize - 1;
+        for (int i = 0; i < psSize; i++) {
+            padded[k--] = (byte)0xff;
+        }
+    }
+
+    public static void main(String argv[]) {
+        int bounds = 1024;
+        int lim = 500000;
+        long start = System.currentTimeMillis();
+        for (int j = 0; j < lim; j++) {
+            paddedSize = j % bounds;
+            padV15(new byte[paddedSize]);
+        }
+        long end = System.currentTimeMillis();
+        System.out.println(end - start);
+        start = System.currentTimeMillis();
+        for (int j = 0; j < lim; j++) {
+            paddedSize = j % bounds;
+            padV15_2(paddedSize);
+        }
+        end = System.currentTimeMillis();
+        System.out.println(end - start);
+        start = System.currentTimeMillis();
+        for (int j = 0; j < lim; j++) {
+            paddedSize = j % bounds;
+            padV15_3();
+        }
+        end = System.currentTimeMillis();
+        System.out.println(end - start);
+        start = System.currentTimeMillis();
+        for (int j = 0; j < lim; j++) {
+            paddedSize = j % bounds;
+            padV15_4();
+        }
+        end = System.currentTimeMillis();
+        System.out.println(end - start);
+        start = System.currentTimeMillis();
+        for (int j = 0; j < lim; j++) {
+            paddedSize = j % bounds;
+            padV15_5();
+        }
+        end = System.currentTimeMillis();
+        System.out.println(end - start);
+    }
+}