diff --git a/design_dump.cc b/design_dump.cc
index cc7f0f293..5f1dcd91b 100644
--- a/design_dump.cc
+++ b/design_dump.cc
@@ -718,6 +718,11 @@ void NetProcTop::dump(ostream&o, unsigned ind) const
       statement_->dump(o, ind+2);
 }
 
+void NetAlloc::dump(ostream&o, unsigned ind) const
+{
+      o << setw(ind) << "// allocate storage : " << scope_path(scope_) << endl;
+}
+
 void NetAssign_::dump_lval(ostream&o) const
 {
       if (sig_) {
@@ -950,6 +955,11 @@ void NetForever::dump(ostream&o, unsigned ind) const
       statement_->dump(o, ind+2);
 }
 
+void NetFree::dump(ostream&o, unsigned ind) const
+{
+      o << setw(ind) << "// free storage : " << scope_path(scope_) << endl;
+}
+
 void NetFuncDef::dump(ostream&o, unsigned ind) const
 {
       o << setw(ind) << "" << "function definition for " << scope_path(scope_) << endl;
diff --git a/elaborate.cc b/elaborate.cc
index bd8aa89f3..add19d661 100644
--- a/elaborate.cc
+++ b/elaborate.cc
@@ -2408,6 +2408,13 @@ NetProc* PCallTask::elaborate_usr(Design*des, NetScope*scope) const
 		  return block;
       }
 
+        /* If this is an automatic task, generate a statement to
+           allocate the local storage. */
+      if (task->is_auto()) {
+	    NetAlloc*ap = new NetAlloc(task);
+	    block->append(ap);
+      }
+
 	/* Generate assignment statement statements for the input and
 	   INOUT ports of the task. These are managed by writing
 	   assignments with the task port the l-value and the passed
@@ -2487,6 +2494,13 @@ NetProc* PCallTask::elaborate_usr(Design*des, NetScope*scope) const
 	    block->append(ass);
       }
 
+        /* If this is an automatic task, generate a statement to free
+           the local storage. */
+      if (task->is_auto()) {
+	    NetFree*fp = new NetFree(task);
+	    block->append(fp);
+      }
+
       return block;
 }
 
diff --git a/emit.cc b/emit.cc
index 1370cb617..6d1be63a8 100644
--- a/emit.cc
+++ b/emit.cc
@@ -196,6 +196,12 @@ bool NetProc::emit_proc(struct target_t*tgt) const
       return false;
 }
 
+bool NetAlloc::emit_proc(struct target_t*tgt) const
+{
+      tgt->proc_alloc(this);
+      return true;
+}
+
 bool NetAssign::emit_proc(struct target_t*tgt) const
 {
       return tgt->proc_assign(this);
@@ -249,6 +255,12 @@ bool NetForever::emit_proc(struct target_t*tgt) const
       return true;
 }
 
+bool NetFree::emit_proc(struct target_t*tgt) const
+{
+      tgt->proc_free(this);
+      return true;
+}
+
 bool NetPDelay::emit_proc(struct target_t*tgt) const
 {
       return tgt->proc_delay(this);
diff --git a/ivl_target.h b/ivl_target.h
index 159ff7dac..021d57249 100644
--- a/ivl_target.h
+++ b/ivl_target.h
@@ -343,6 +343,7 @@ typedef enum ivl_signal_type_e {
 typedef enum ivl_statement_type_e {
       IVL_ST_NONE    = 0,
       IVL_ST_NOOP    = 1,
+      IVL_ST_ALLOC   = 25,
       IVL_ST_ASSIGN    = 2,
       IVL_ST_ASSIGN_NB = 3,
       IVL_ST_BLOCK   = 4,
@@ -359,6 +360,7 @@ typedef enum ivl_statement_type_e {
       IVL_ST_FORCE   = 14,
       IVL_ST_FOREVER = 15,
       IVL_ST_FORK    = 16,
+      IVL_ST_FREE    = 26,
       IVL_ST_RELEASE = 17,
       IVL_ST_REPEAT  = 18,
       IVL_ST_STASK   = 19,
diff --git a/netlist.cc b/netlist.cc
index b4ce2697c..60d8ccd4c 100644
--- a/netlist.cc
+++ b/netlist.cc
@@ -1973,6 +1973,44 @@ const NetScope* NetUTask::task() const
       return task_;
 }
 
+NetAlloc::NetAlloc(NetScope*scope)
+: scope_(scope)
+{
+}
+
+NetAlloc::~NetAlloc()
+{
+}
+#if 0
+const string NetAlloc::name() const
+{
+      return scope_->name();
+}
+#endif
+const NetScope* NetAlloc::scope() const
+{
+      return scope_;
+}
+
+NetFree::NetFree(NetScope*scope)
+: scope_(scope)
+{
+}
+
+NetFree::~NetFree()
+{
+}
+#if 0
+const string NetFree::name() const
+{
+      return scope_->name();
+}
+#endif
+const NetScope* NetFree::scope() const
+{
+      return scope_;
+}
+
 NetExpr::NetExpr(unsigned w)
 : width_(w), signed_flag_(false)
 {
diff --git a/netlist.h b/netlist.h
index fc2d120f7..f0d2407c9 100644
--- a/netlist.h
+++ b/netlist.h
@@ -2086,6 +2086,23 @@ class NetProc : public virtual LineInfo {
       NetProc& operator= (const NetProc&);
 };
 
+class NetAlloc  : public NetProc {
+
+    public:
+      NetAlloc(NetScope*);
+      ~NetAlloc();
+
+      const string name() const;
+
+      const NetScope* scope() const;
+
+      virtual bool emit_proc(struct target_t*) const;
+      virtual void dump(ostream&, unsigned ind) const;
+
+    private:
+      NetScope*scope_;
+};
+
 /*
  * Procedural assignment is broken into a suite of classes. These
  * classes represent the various aspects of the assignment statement
@@ -2705,6 +2722,23 @@ class NetForever : public NetProc {
       NetProc*statement_;
 };
 
+class NetFree   : public NetProc {
+
+    public:
+      NetFree(NetScope*);
+      ~NetFree();
+
+      const string name() const;
+
+      const NetScope* scope() const;
+
+      virtual bool emit_proc(struct target_t*) const;
+      virtual void dump(ostream&, unsigned ind) const;
+
+    private:
+      NetScope*scope_;
+};
+
 /*
  * A function definition is elaborated just like a task, though by now
  * it is certain that the first parameter (a phantom parameter) is the
diff --git a/parse.y b/parse.y
index 75bcf60dc..bbd049151 100644
--- a/parse.y
+++ b/parse.y
@@ -459,10 +459,10 @@ block_item_decl
      with real value. Note that real and realtime are interchangeable
      in this context. */
 
-  | attribute_list_opt K_real real_variable_list ';'
-      { delete $3; }
-  | attribute_list_opt K_realtime real_variable_list ';'
-      { delete $3; }
+        | attribute_list_opt K_real real_variable_list ';'
+                { delete $3; }
+        | attribute_list_opt K_realtime real_variable_list ';'
+                { delete $3; }
 
 	| K_event list_of_identifiers ';'
 		{ pform_make_events($2, @1.text, @1.first_line);
diff --git a/t-dll-api.cc b/t-dll-api.cc
index c68a61f32..ebfee98e5 100644
--- a/t-dll-api.cc
+++ b/t-dll-api.cc
@@ -1913,9 +1913,15 @@ extern "C" ivl_statement_t ivl_stmt_block_stmt(ivl_statement_t net,
 extern "C" ivl_scope_t ivl_stmt_call(ivl_statement_t net)
 {
       switch (net->type_) {
+	  case IVL_ST_ALLOC:
+	    return net->u_.alloc_.scope;
+
 	  case IVL_ST_DISABLE:
 	    return net->u_.disable_.scope;
 
+	  case IVL_ST_FREE:
+	    return net->u_.free_.scope;
+
 	  case IVL_ST_UTASK:
 	    return net->u_.utask_.def;
 	  default:
diff --git a/t-dll-proc.cc b/t-dll-proc.cc
index badf7cd78..b3b96973f 100644
--- a/t-dll-proc.cc
+++ b/t-dll-proc.cc
@@ -188,6 +188,16 @@ void dll_target::make_assign_lvals_(const NetAssignBase*net)
       }
 }
 
+void dll_target::proc_alloc(const NetAlloc*net)
+{
+      assert(stmt_cur_);
+      assert(stmt_cur_->type_ == IVL_ST_NONE);
+      FILE_NAME(stmt_cur_, net);
+
+      stmt_cur_->type_ = IVL_ST_ALLOC;
+      stmt_cur_->u_.alloc_.scope = lookup_scope_(net->scope());
+}
+
 /*
  */
 bool dll_target::proc_assign(const NetAssign*net)
@@ -629,6 +639,16 @@ void dll_target::proc_forever(const NetForever*net)
       stmt_cur_ = save_cur_;
 }
 
+void dll_target::proc_free(const NetFree*net)
+{
+      assert(stmt_cur_);
+      assert(stmt_cur_->type_ == IVL_ST_NONE);
+      FILE_NAME(stmt_cur_, net);
+
+      stmt_cur_->type_ = IVL_ST_FREE;
+      stmt_cur_->u_.free_.scope = lookup_scope_(net->scope());
+}
+
 bool dll_target::proc_release(const NetRelease*net)
 {
       assert(stmt_cur_);
diff --git a/t-dll.h b/t-dll.h
index 87047b9d4..7989ab8f9 100644
--- a/t-dll.h
+++ b/t-dll.h
@@ -115,6 +115,7 @@ struct dll_target  : public target_t, public expr_scan_t {
 	/* These methods and members are used for forming the
 	   statements of a thread. */
       struct ivl_statement_s*stmt_cur_;
+      void proc_alloc(const NetAlloc*);
       bool proc_assign(const NetAssign*);
       void proc_assign_nb(const NetAssignNB*);
       bool proc_block(const NetBlock*);
@@ -126,6 +127,7 @@ struct dll_target  : public target_t, public expr_scan_t {
       bool proc_disable(const NetDisable*);
       bool proc_force(const NetForce*);
       void proc_forever(const NetForever*);
+      void proc_free(const NetFree*);
       bool proc_release(const NetRelease*);
       void proc_repeat(const NetRepeat*);
       void proc_stask(const NetSTask*);
@@ -665,6 +667,10 @@ struct ivl_statement_s {
       unsigned lineno;
 
       union {
+	    struct { /* IVL_ST_ALLOC */
+		  ivl_scope_t scope;
+	    } alloc_;
+
 	    struct { /* IVL_ST_ASSIGN IVL_ST_ASSIGN_NB
 			IVL_ST_CASSIGN, IVL_ST_DEASSIGN */
 		  unsigned lvals_;
@@ -718,6 +724,10 @@ struct ivl_statement_s {
 		  ivl_statement_t stmt_;
 	    } forever_;
 
+	    struct { /* IVL_ST_FREE */
+		  ivl_scope_t scope;
+	    } free_;
+
 	    struct { /* IVL_ST_STASK */
 		  const char*name_;
 		  unsigned   nparm_;
diff --git a/target.cc b/target.cc
index fef82cedb..00bd9d0cb 100644
--- a/target.cc
+++ b/target.cc
@@ -243,6 +243,12 @@ bool target_t::process(const NetProcTop*top)
       return top->statement()->emit_proc(this);
 }
 
+void target_t::proc_alloc(const NetAlloc*)
+{
+      cerr << "target (" << typeid(*this).name() <<  "): "
+	    "Unhandled proc_alloc." << endl;
+}
+
 bool target_t::proc_assign(const NetAssign*)
 {
       cerr << "target (" << typeid(*this).name() <<  "): "
@@ -322,6 +328,12 @@ void target_t::proc_forever(const NetForever*)
 	    "Unhandled proc_forever." << endl;
 }
 
+void target_t::proc_free(const NetFree*)
+{
+      cerr << "target (" << typeid(*this).name() <<  "): "
+	    "Unhandled proc_free." << endl;
+}
+
 bool target_t::proc_release(const NetRelease*dev)
 {
       cerr << dev->get_fileline() << ": internal error: "
diff --git a/target.h b/target.h
index a266d6d15..2be4acf20 100644
--- a/target.h
+++ b/target.h
@@ -105,6 +105,7 @@ struct target_t {
       virtual bool process(const NetProcTop*);
 
 	/* Various kinds of process nodes are dispatched through these. */
+      virtual void proc_alloc(const NetAlloc*);
       virtual bool proc_assign(const NetAssign*);
       virtual void proc_assign_nb(const NetAssignNB*);
       virtual bool proc_block(const NetBlock*);
@@ -116,6 +117,7 @@ struct target_t {
       virtual bool proc_disable(const NetDisable*);
       virtual bool proc_force(const NetForce*);
       virtual void proc_forever(const NetForever*);
+      virtual void proc_free(const NetFree*);
       virtual bool proc_release(const NetRelease*);
       virtual void proc_repeat(const NetRepeat*);
       virtual bool proc_trigger(const NetEvTrig*);
diff --git a/tgt-stub/statement.c b/tgt-stub/statement.c
index bf560cf5f..b7807bbcb 100644
--- a/tgt-stub/statement.c
+++ b/tgt-stub/statement.c
@@ -183,6 +183,10 @@ void show_statement(ivl_statement_t net, unsigned ind)
 
       switch (code) {
 
+	  case IVL_ST_ALLOC:
+	    fprintf(out, "%*sallocate automatic storage ...\n", ind, "");
+            break;
+
 	  case IVL_ST_ASSIGN:
 	    fprintf(out, "%*sASSIGN <lwidth=%u>\n", ind, "",
 		    ivl_stmt_lwidth(net));
@@ -316,6 +320,10 @@ void show_statement(ivl_statement_t net, unsigned ind)
 		break;
 	  }
 
+	  case IVL_ST_FREE:
+	    fprintf(out, "%*sfree automatic storage ...\n", ind, "");
+            break;
+
 	  case IVL_ST_NOOP:
 	    fprintf(out, "%*s/* noop */;\n", ind, "");
 	    break;
@@ -356,4 +364,3 @@ void show_statement(ivl_statement_t net, unsigned ind)
 	    fprintf(out, "%*sunknown statement type (%u)\n", ind, "", code);
       }
 }
-
diff --git a/tgt-vvp/draw_ufunc.c b/tgt-vvp/draw_ufunc.c
index 6ea3b3637..5a1cfcf44 100644
--- a/tgt-vvp/draw_ufunc.c
+++ b/tgt-vvp/draw_ufunc.c
@@ -88,6 +88,11 @@ struct vector_info draw_ufunc_expr(ivl_expr_t exp, unsigned wid)
       ivl_signal_t retval = ivl_scope_port(def, 0);
       struct vector_info res;
 
+        /* If this is an automatic function, allocate the local storage. */
+      if (ivl_scope_is_auto(def)) {
+            fprintf(vvp_out, "    %%alloc S_%p;\n", def);
+      }
+
 	/* evaluate the expressions and send the results to the
 	   function ports. */
 
@@ -134,6 +139,11 @@ struct vector_info draw_ufunc_expr(ivl_expr_t exp, unsigned wid)
       if (load_wid < wid)
 	    pad_expr_in_place(exp, res, swid);
 
+        /* If this is an automatic function, free the local storage. */
+      if (ivl_scope_is_auto(def)) {
+            fprintf(vvp_out, "    %%free S_%p;\n", def);
+      }
+
       return res;
 }
 
@@ -144,6 +154,11 @@ int draw_ufunc_real(ivl_expr_t exp)
       int res = 0;
       int idx;
 
+        /* If this is an automatic function, allocate the local storage. */
+      if (ivl_scope_is_auto(def)) {
+            fprintf(vvp_out, "    %%alloc S_%p;\n", def);
+      }
+
       assert(ivl_expr_parms(exp) == (ivl_scope_ports(def)-1));
       for (idx = 0 ;  idx < ivl_expr_parms(exp) ;  idx += 1) {
 	    ivl_signal_t port = ivl_scope_port(def, idx+1);
@@ -163,6 +178,10 @@ int draw_ufunc_real(ivl_expr_t exp)
       res = allocate_word();
       fprintf(vvp_out, "  %%load/wr %d, v%p_0;\n", res, retval);
 
+        /* If this is an automatic function, free the local storage. */
+      if (ivl_scope_is_auto(def)) {
+            fprintf(vvp_out, "    %%free S_%p;\n", def);
+      }
+
       return res;
 }
-
diff --git a/tgt-vvp/vvp_process.c b/tgt-vvp/vvp_process.c
index e6233a363..e23143aa4 100644
--- a/tgt-vvp/vvp_process.c
+++ b/tgt-vvp/vvp_process.c
@@ -428,6 +428,14 @@ static void set_vec_to_lval(ivl_statement_t net, struct vector_info res)
       }
 }
 
+static int show_stmt_alloc(ivl_statement_t net)
+{
+      ivl_scope_t scope = ivl_stmt_call(net);
+
+      fprintf(vvp_out, "    %%alloc S_%p;\n", scope);
+      return 0;
+}
+
 static int show_stmt_assign_vector(ivl_statement_t net)
 {
       ivl_expr_t rval = ivl_stmt_rval(net);
@@ -1398,6 +1406,14 @@ static int show_stmt_fork(ivl_statement_t net, ivl_scope_t sscope)
       return rc;
 }
 
+static int show_stmt_free(ivl_statement_t net)
+{
+      ivl_scope_t scope = ivl_stmt_call(net);
+
+      fprintf(vvp_out, "    %%free S_%p;\n", scope);
+      return 0;
+}
+
 /*
  * noop statements are implemented by doing nothing.
  */
@@ -1645,6 +1661,10 @@ static int show_statement(ivl_statement_t net, ivl_scope_t sscope)
 
       switch (code) {
 
+	  case IVL_ST_ALLOC:
+	    rc += show_stmt_alloc(net);
+	    break;
+
 	  case IVL_ST_ASSIGN:
 	    rc += show_stmt_assign(net);
 	    break;
@@ -1706,6 +1726,10 @@ static int show_statement(ivl_statement_t net, ivl_scope_t sscope)
 	    rc += show_stmt_fork(net, sscope);
 	    break;
 
+	  case IVL_ST_FREE:
+	    rc += show_stmt_free(net);
+	    break;
+
 	  case IVL_ST_NOOP:
 	    rc += show_stmt_noop(net);
 	    break;
diff --git a/tgt-vvp/vvp_scope.c b/tgt-vvp/vvp_scope.c
index 32780a41d..75f36fc9c 100644
--- a/tgt-vvp/vvp_scope.c
+++ b/tgt-vvp/vvp_scope.c
@@ -1581,7 +1581,7 @@ static void draw_lpm_ufunc(ivl_lpm_t net)
 
       fprintf(vvp_out, ")");
 
-	/* Finally, print the reference to the signal from which the
+	/* Now print the reference to the signal from which the
 	   result is collected. */
       { ivl_signal_t psig = ivl_scope_port(def, 0);
         assert(ivl_lpm_width(net) == ivl_signal_width(psig));
@@ -1590,7 +1590,8 @@ static void draw_lpm_ufunc(ivl_lpm_t net)
 	fprintf(vvp_out, " v%p_0", psig);
       }
 
-      fprintf(vvp_out, ";\n");
+        /* Finally, print the scope identifier. */
+      fprintf(vvp_out, " S_%p;\n", def);
 }
 
 /*
@@ -1774,13 +1775,9 @@ int draw_scope(ivl_scope_t net, ivl_scope_t parent)
 {
       unsigned idx;
       const char *type;
-	/* For now we do not support automatic tasks or functions. */
-      if (ivl_scope_is_auto(net)) {
-	    fprintf(stderr, "%s:%u: vvp-tgt sorry: automatic tasks/functions "
-	                    "are not supported!\n",
-	                    ivl_scope_def_file(net), ivl_scope_def_lineno(net));
-	    exit(1);
-      }
+
+      const char*prefix = ivl_scope_is_auto(net) ? "auto" : "";
+
       switch (ivl_scope_type(net)) {
       case IVL_SCT_MODULE:   type = "module";   break;
       case IVL_SCT_FUNCTION: type = "function"; break;
@@ -1791,8 +1788,8 @@ int draw_scope(ivl_scope_t net, ivl_scope_t parent)
       default:               type = "?";        assert(0);
       }
 
-      fprintf(vvp_out, "S_%p .scope %s, \"%s\" \"%s\" %d %d",
-	      net, type, vvp_mangle_name(ivl_scope_basename(net)),
+      fprintf(vvp_out, "S_%p .scope %s%s, \"%s\" \"%s\" %d %d",
+	      net, prefix, type, vvp_mangle_name(ivl_scope_basename(net)),
               ivl_scope_tname(net), ivl_file_table_index(ivl_scope_file(net)),
               ivl_scope_lineno(net));
 
diff --git a/vvp/README.txt b/vvp/README.txt
index 188f58e36..4781f6814 100644
--- a/vvp/README.txt
+++ b/vvp/README.txt
@@ -96,8 +96,8 @@ The syntax of a scope statement is:
 
         <label> .scope <type>, <instance>, <declaration>, <parent> ;
 
-The <type> is the general type of the scope: module, function, task,
-begin, fork or generate.
+The <type> is the general type of the scope: module, autofunction,
+function, autotask, task, begin, fork or generate.
 
 The <instance> is a string that is the base name of the instance. For
 modules, this is the instance name. For tasks, this is the task name.
@@ -656,7 +656,7 @@ STRUCTURAL FUNCTION CALLS:
 
 The .ufunc statement defines a call to a user defined function.
 
-	<label> .ufunc <flabel>, <wid>, <isymbols> ( <psymbols> ) <rsymbol> ;
+	<label> .ufunc <flabel>, <wid>, <isymbols> ( <psymbols> ) <rsymbol> <ssymbol>;
 
 The <flabel> is the code label for the first instruction of the
 function implementation. This is code that the simulator will branch
@@ -673,9 +673,11 @@ list. The <psymbols> are variables that represent the input ports for
 the function. The ufunc performs an assignment to these variables
 before calling the function.
 
-Finally, the <rsymbol> is the variable within the function where the
-result will be found when the function code ends. This value is picked
-up and propagated to the output of the functor.
+The <rsymbol> is the variable within the function where the result
+will be found when the function code ends. This value is picked up
+and propagated to the output of the functor.
+
+The <ssymbol> is the function scope name.
 
 THREAD STATEMENTS:
 
@@ -1035,6 +1037,60 @@ This creates a functor and makes it into a mode-2 functor. Then the
 trigger statement, "-> a", cause a ``%set a, 0;'' statement be
 generated. This is sufficient to trigger the event.
 
+
+AUTOMATICALLY ALLOCATED SCOPES
+
+If a .scope statement has a <type> of autofunction or autotask, the
+scope is flagged as being an automatically allocated scope. The functor
+for each variable or event declared in that scope is added to a list
+of items that need to be automatically allocated for each dynamic
+instance of that scope.
+
+Before copying the input parameters of an automatic function or task
+into the scope variables, a new scope instance needs to be allocated.
+For function or task calls in procedural code, this is handled by the
+%alloc instruction. For structural function calls, this is handled
+by the phantom code generated by the .ufunc statement. In both cases,
+VVP attempts to use a previously freed scope instance - only if none
+are available is a new instance created.
+
+After copying the result of an automatic function or the output
+parameters of an automatic task, the scope instance can be freed.
+For function or task calls in procedural code, this is handled by the
+%free instruction. For structural function calls, this is handled
+by the phantom code generated by the .ufunc statement. In both cases,
+VVP adds the instance to a list of freed instances for that scope,
+which allows the storage to be reused the next time a new instance
+is required.
+
+For each automatically allocated scope instance, VVP creates an array
+of items, referred to as the scope context. Each item in this array is
+a pointer to the allocated storage for holding the state of one scope
+variable or event. The index into this array for any given variable
+or event, referred to as the context index, is stored in the functor
+associated with that variable or event.
+
+Each VVP thread keeps track of its current write context and current
+read context. For threads executing in a static scope, these are both
+initialised to null values. For threads executing in an automatically
+allocated scope, these are both initialised to refer to the context
+allocated to that scope.
+
+Before starting the copying of the input parameters of an automatic
+function or task, the current write context of the caller thread is
+set to the context allocated for that function/task call. After the
+thread that executed the function/task body has been rejoined and
+before starting the copying of the result or output parameters, the
+current write context is reset to its previous value and the current
+read context is set to the context allocated for the function/task
+call. After finishing the copying of the result or output parameters,
+the current read context is reset to its previous value.
+
+When reading or writing the state of an automatically allocated
+variable or event, the associated functor indirects through the
+current read or write context of the running thread, using its
+stored context index.
+
 /*
  * Copyright (c) 2001 Stephen Williams (steve@icarus.com)
  *
diff --git a/vvp/array.cc b/vvp/array.cc
index eb3454182..936ca9c12 100644
--- a/vvp/array.cc
+++ b/vvp/array.cc
@@ -852,7 +852,9 @@ void compile_var_array(char*label, char*name, int last, int first,
 
 	/* Make the words. */
       arr->vals_width = labs(msb-lsb) + 1;
-      arr->vals = new vvp_vector4array_t(arr->vals_width, arr->array_count);
+      arr->vals = new vvp_vector4array_t(arr->vals_width, arr->array_count,
+                                         vpip_peek_current_scope()->is_automatic);
+      vpip_add_item_to_current_scope(arr->vals);
       vpip_make_dec_const(&arr->msb, msb);
       vpip_make_dec_const(&arr->lsb, lsb);
 
diff --git a/vvp/codes.h b/vvp/codes.h
index 28e600c2c..769b3a915 100644
--- a/vvp/codes.h
+++ b/vvp/codes.h
@@ -36,6 +36,7 @@ extern bool of_ABS_WR(vthread_t thr, vvp_code_t code);
 extern bool of_ADD(vthread_t thr, vvp_code_t code);
 extern bool of_ADD_WR(vthread_t thr, vvp_code_t code);
 extern bool of_ADDI(vthread_t thr, vvp_code_t code);
+extern bool of_ALLOC(vthread_t thr, vvp_code_t code);
 extern bool of_AND(vthread_t thr, vvp_code_t code);
 extern bool of_ANDI(vthread_t thr, vvp_code_t code);
 extern bool of_ANDR(vthread_t thr, vvp_code_t code);
@@ -91,6 +92,7 @@ extern bool of_FORCE_V(vthread_t thr, vvp_code_t code);
 extern bool of_FORCE_WR(vthread_t thr, vvp_code_t code);
 extern bool of_FORCE_X0(vthread_t thr, vvp_code_t code);
 extern bool of_FORK(vthread_t thr, vvp_code_t code);
+extern bool of_FREE(vthread_t thr, vvp_code_t code);
 extern bool of_INV(vthread_t thr, vvp_code_t code);
 extern bool of_IX_ADD(vthread_t thr, vvp_code_t code);
 extern bool of_IX_GET(vthread_t thr, vvp_code_t code);
@@ -156,8 +158,7 @@ extern bool of_XORR(vthread_t thr, vvp_code_t code);
 
 extern bool of_ZOMBIE(vthread_t thr, vvp_code_t code);
 
-extern bool of_FORK_UFUNC(vthread_t thr, vvp_code_t code);
-extern bool of_JOIN_UFUNC(vthread_t thr, vvp_code_t code);
+extern bool of_EXEC_UFUNC(vthread_t thr, vvp_code_t code);
 
 extern bool of_CHUNK_LINK(vthread_t thr, vvp_code_t code);
 
diff --git a/vvp/compile.cc b/vvp/compile.cc
index 64a06fc57..672b6e08e 100644
--- a/vvp/compile.cc
+++ b/vvp/compile.cc
@@ -1687,6 +1687,34 @@ void compile_fork(char*label, struct symb_s dest, struct symb_s scope)
       compile_vpi_lookup(&code->handle, scope.text);
 }
 
+void compile_alloc(char*label, struct symb_s scope)
+{
+      if (label)
+	    compile_codelabel(label);
+
+
+	/* Fill in the basics of the %alloc in the instruction. */
+      vvp_code_t code = codespace_allocate();
+      code->opcode = of_ALLOC;
+
+	/* Figure out the target SCOPE. */
+      compile_vpi_lookup(&code->handle, scope.text);
+}
+
+void compile_free(char*label, struct symb_s scope)
+{
+      if (label)
+	    compile_codelabel(label);
+
+
+	/* Fill in the basics of the %free in the instruction. */
+      vvp_code_t code = codespace_allocate();
+      code->opcode = of_FREE;
+
+	/* Figure out the target SCOPE. */
+      compile_vpi_lookup(&code->handle, scope.text);
+}
+
 void compile_vpi_call(char*label, char*name,
                       long file_idx, long lineno,
                       unsigned argc, vpiHandle*argv)
diff --git a/vvp/compile.h b/vvp/compile.h
index 989e7747d..e816a24a4 100644
--- a/vvp/compile.h
+++ b/vvp/compile.h
@@ -347,7 +347,7 @@ extern void compile_array_cleanup(void);
 extern void compile_ufunc(char*label, char*code, unsigned wid,
 			  unsigned argc, struct symb_s*argv,
 			  unsigned portc, struct symb_s*portv,
-			  struct symb_s retv);
+			  struct symb_s retv, struct symb_s scope);
 
 /*
  * The compile_event function takes the parts of the event statement
@@ -406,6 +406,9 @@ extern void compile_fork(char*label, struct symb_s targ_s,
 			 struct symb_s scope_s);
 extern void compile_codelabel(char*label);
 
+extern void compile_alloc(char*label, struct symb_s scope_s);
+extern void compile_free(char*label, struct symb_s scope_s);
+
 /*
  * The parser uses these functions to compile .scope statements.
  * The implementations of these live in the vpi_scope.cc file.
diff --git a/vvp/event.cc b/vvp/event.cc
index 0f80a6efa..91e2c306a 100644
--- a/vvp/event.cc
+++ b/vvp/event.cc
@@ -32,7 +32,7 @@
 
 # include <iostream>
 
-void waitable_hooks_s::run_waiting_threads_()
+void waitable_hooks_s::run_waiting_threads_(unsigned context_idx)
 {
 	// Run the non-blocking event controls.
       last = &event_ctls;
@@ -48,12 +48,17 @@ void waitable_hooks_s::run_waiting_threads_()
 	    }
       }
 
-      if (threads == 0)
-	    return;
-
-      vthread_t tmp = threads;
-      threads = 0;
-      vthread_schedule_list(tmp);
+      vthread_t tmp;
+      if (context_idx) {
+            waitable_state_s*state = static_cast<waitable_state_s*>
+                  (vthread_get_wt_context_item(context_idx));
+            tmp = state->threads;
+            state->threads = 0;
+      } else {
+            tmp = threads;
+            threads = 0;
+      }
+      if (tmp) vthread_schedule_list(tmp);
 }
 
 evctl::evctl(unsigned long ecount)
@@ -180,6 +185,12 @@ const vvp_fun_edge::edge_t vvp_edge_negedge
 
 const vvp_fun_edge::edge_t vvp_edge_none    = 0;
 
+struct vvp_fun_edge_state_s : public waitable_state_s {
+      vvp_fun_edge_state_s() : bit(BIT4_X) {}
+
+      vvp_bit4_t bit;
+};
+
 vvp_fun_edge::vvp_fun_edge(edge_t e, bool debug_flag)
 : edge_(e), debug_(debug_flag)
 {
@@ -193,16 +204,38 @@ vvp_fun_edge::~vvp_fun_edge()
 {
 }
 
+void vvp_fun_edge::alloc_instance(vvp_context_t context)
+{
+      vvp_set_context_item(context, context_idx, new vvp_fun_edge_state_s);
+}
+
+void vvp_fun_edge::reset_instance(vvp_context_t context)
+{
+      vvp_fun_edge_state_s*state = static_cast<vvp_fun_edge_state_s*>
+            (vvp_get_context_item(context, context_idx));
+      state->threads = 0;
+      state->bit = BIT4_X;
+}
+
 void vvp_fun_edge::recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit)
 {
+      vvp_bit4_t*old_bit;
+      if (context_idx) {
+            vvp_fun_edge_state_s*state = static_cast<vvp_fun_edge_state_s*>
+                  (vthread_get_wt_context_item(context_idx));
+            old_bit = &state->bit;
+      } else {
+            old_bit = &bits_[port.port()];
+      }
+
 	/* See what kind of edge this represents. */
-      edge_t mask = VVP_EDGE(bits_[port.port()], bit.value(0));
+      edge_t mask = VVP_EDGE(*old_bit, bit.value(0));
 
 	/* Save the current input for the next time around. */
-      bits_[port.port()] = bit.value(0);
+      *old_bit = bit.value(0);
 
       if ((edge_ == vvp_edge_none) || (edge_ & mask)) {
-	    run_waiting_threads_();
+	    run_waiting_threads_(context_idx);
 
 	    vvp_net_t*net = port.ptr();
 	    vvp_send_vec4(net->out, bit);
@@ -210,6 +243,13 @@ void vvp_fun_edge::recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit)
 }
 
 
+struct vvp_fun_anyedge_state_s : public waitable_state_s {
+      vvp_fun_anyedge_state_s() : bitsr(0.0) {}
+
+      vvp_vector4_t bits;
+      double bitsr;
+};
+
 vvp_fun_anyedge::vvp_fun_anyedge(bool debug_flag)
 : debug_(debug_flag)
 {
@@ -221,17 +261,39 @@ vvp_fun_anyedge::~vvp_fun_anyedge()
 {
 }
 
+void vvp_fun_anyedge::alloc_instance(vvp_context_t context)
+{
+      vvp_set_context_item(context, context_idx, new vvp_fun_anyedge_state_s);
+}
+
+void vvp_fun_anyedge::reset_instance(vvp_context_t context)
+{
+      vvp_fun_anyedge_state_s*state = static_cast<vvp_fun_anyedge_state_s*>
+            (vvp_get_context_item(context, context_idx));
+      state->threads = 0;
+      state->bits.set_to_x();
+      state->bitsr = 0.0;
+}
+
 void vvp_fun_anyedge::recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit)
 {
-      unsigned pdx = port.port();
       bool flag = false;
 
-      if (bits_[pdx].size() != bit.size()) {
+      vvp_vector4_t*old_bits;
+      if (context_idx) {
+            vvp_fun_anyedge_state_s*state = static_cast<vvp_fun_anyedge_state_s*>
+                  (vthread_get_wt_context_item(context_idx));
+            old_bits = &state->bits;
+      } else {
+            old_bits = &bits_[port.port()];
+      }
+
+      if (old_bits->size() != bit.size()) {
 	    flag = true;
 
       } else {
 	    for (unsigned idx = 0 ;  idx < bit.size() ;  idx += 1) {
-		  if (bits_[pdx].value(idx) != bit.value(idx)) {
+		  if (old_bits->value(idx) != bit.value(idx)) {
 			flag = true;
 			break;
 		  }
@@ -239,8 +301,8 @@ void vvp_fun_anyedge::recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit)
       }
 
       if (flag) {
-	    bits_[pdx] = bit;
-	    run_waiting_threads_();
+	    *old_bits = bit;
+	    run_waiting_threads_(context_idx);
 	    vvp_net_t*net = port.ptr();
 	    vvp_send_vec4(net->out, bit);
       }
@@ -248,16 +310,18 @@ void vvp_fun_anyedge::recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit)
 
 void vvp_fun_anyedge::recv_real(vvp_net_ptr_t port, double bit)
 {
-      unsigned pdx = port.port();
-      bool flag = false;
-
-      if (bitsr_[pdx] != bit) {
-	    flag = true;
-	    bitsr_[pdx] = bit;
+      double*old_bits;
+      if (context_idx) {
+            vvp_fun_anyedge_state_s*state = static_cast<vvp_fun_anyedge_state_s*>
+                  (vthread_get_wt_context_item(context_idx));
+            old_bits = &state->bitsr;
+      } else {
+            old_bits = &bitsr_[port.port()];
       }
 
-      if (flag) {
-	    run_waiting_threads_();
+      if (*old_bits != bit) {
+	    *old_bits = bit;
+	    run_waiting_threads_(context_idx);
 	    vvp_net_t*net = port.ptr();
 	    vvp_send_vec4(net->out, vvp_vector4_t());
       }
@@ -271,9 +335,21 @@ vvp_fun_event_or::~vvp_fun_event_or()
 {
 }
 
+void vvp_fun_event_or::alloc_instance(vvp_context_t context)
+{
+      vvp_set_context_item(context, context_idx, new waitable_state_s);
+}
+
+void vvp_fun_event_or::reset_instance(vvp_context_t context)
+{
+      waitable_state_s*state = static_cast<waitable_state_s*>
+            (vvp_get_context_item(context, context_idx));
+      state->threads = 0;
+}
+
 void vvp_fun_event_or::recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit)
 {
-      run_waiting_threads_();
+      run_waiting_threads_(context_idx);
       vvp_net_t*net = port.ptr();
       vvp_send_vec4(net->out, bit);
 }
@@ -287,9 +363,21 @@ vvp_named_event::~vvp_named_event()
 {
 }
 
+void vvp_named_event::alloc_instance(vvp_context_t context)
+{
+      vvp_set_context_item(context, context_idx, new waitable_state_s);
+}
+
+void vvp_named_event::reset_instance(vvp_context_t context)
+{
+      waitable_state_s*state = static_cast<waitable_state_s*>
+            (vvp_get_context_item(context, context_idx));
+      state->threads = 0;
+}
+
 void vvp_named_event::recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit)
 {
-      run_waiting_threads_();
+      run_waiting_threads_(context_idx);
       vvp_net_t*net = port.ptr();
       vvp_send_vec4(net->out, bit);
 
@@ -320,7 +408,9 @@ void compile_event(char*label, char*type,
       if (strcmp(type,"edge") == 0) {
 
 	    free(type);
-	    fun = new vvp_fun_anyedge(debug_flag);
+	    vvp_fun_anyedge*event_fun = new vvp_fun_anyedge(debug_flag);
+            vpip_add_item_to_current_scope(event_fun);
+            fun = event_fun;
 
       } else {
 
@@ -334,7 +424,9 @@ void compile_event(char*label, char*type,
 	    assert(argc <= 4);
 	    free(type);
 
-	    fun = new vvp_fun_edge(edge, debug_flag);
+	    vvp_fun_edge*event_fun = new vvp_fun_edge(edge, debug_flag);
+            vpip_add_item_to_current_scope(event_fun);
+            fun = event_fun;
       }
 
       vvp_net_t* ptr = new vvp_net_t;
@@ -348,10 +440,11 @@ void compile_event(char*label, char*type,
 
 static void compile_event_or(char*label, unsigned argc, struct symb_s*argv)
 {
-      vvp_net_fun_t*fun = new vvp_fun_event_or;
+      vvp_fun_event_or*fun = new vvp_fun_event_or;
       vvp_net_t* ptr = new vvp_net_t;
       ptr->fun = fun;
 
+      vpip_add_item_to_current_scope(fun);
       define_functor_symbol(label, ptr);
       free(label);
 
@@ -373,8 +466,10 @@ void compile_named_event(char*label, char*name)
       vvp_net_t*ptr = new vvp_net_t;
 
       vpiHandle obj = vpip_make_named_event(name, ptr);
-      ptr->fun = new vvp_named_event(obj);
+      vvp_named_event*fun = new vvp_named_event(obj);
+      ptr->fun = fun;
 
+      vpip_add_item_to_current_scope(fun);
       define_functor_symbol(label, ptr);
       compile_vpi_symbol(label, obj);
       vpip_attach_to_current_scope(obj);
diff --git a/vvp/event.h b/vvp/event.h
index 26b062dfb..2d66416e3 100644
--- a/vvp/event.h
+++ b/vvp/event.h
@@ -108,7 +108,17 @@ struct waitable_hooks_s {
       evctl**last;
 
     protected:
-      void run_waiting_threads_();
+      void run_waiting_threads_(unsigned context_idx);
+};
+
+/*
+ * This is the base object for storing state information for each instance
+ * of an automatically allocated event. In the general case, all that is
+ * needed is the list of threads waiting on that instance.
+ */
+struct waitable_state_s {
+      waitable_state_s() : threads(0) { }
+      vthread_t threads;
 };
 
 /*
@@ -121,9 +131,11 @@ class vvp_fun_edge : public vvp_net_fun_t, public waitable_hooks_s {
     public:
       typedef unsigned short edge_t;
       explicit vvp_fun_edge(edge_t e, bool debug_flag);
-
       virtual ~vvp_fun_edge();
 
+      void alloc_instance(vvp_context_t context);
+      void reset_instance(vvp_context_t context);
+
       void recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit);
 
     private:
@@ -143,8 +155,8 @@ extern const vvp_fun_edge::edge_t vvp_edge_none;
  * functor looks at the entire input vector for any change.
  *
  * The anyedge is also different in that it can receive real
- * values. in this case, any detectable change in the real value leads
- * to an even trigger.
+ * values. In this case, any detectable change in the real value leads
+ * to an event trigger.
  */
 class vvp_fun_anyedge : public vvp_net_fun_t, public waitable_hooks_s {
 
@@ -152,6 +164,9 @@ class vvp_fun_anyedge : public vvp_net_fun_t, public waitable_hooks_s {
       explicit vvp_fun_anyedge(bool debug_flag);
       virtual ~vvp_fun_anyedge();
 
+      void alloc_instance(vvp_context_t context);
+      void reset_instance(vvp_context_t context);
+
       void recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit);
       void recv_real(vvp_net_ptr_t port, double bit);
 
@@ -172,6 +187,9 @@ class vvp_fun_event_or : public vvp_net_fun_t, public waitable_hooks_s {
       explicit vvp_fun_event_or();
       ~vvp_fun_event_or();
 
+      void alloc_instance(vvp_context_t context);
+      void reset_instance(vvp_context_t context);
+
       void recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit);
 
     private:
@@ -188,6 +206,9 @@ class vvp_named_event : public vvp_net_fun_t, public waitable_hooks_s {
       explicit vvp_named_event(struct __vpiHandle*eh);
       ~vvp_named_event();
 
+      void alloc_instance(vvp_context_t context);
+      void reset_instance(vvp_context_t context);
+
       void recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit);
 
     private:
diff --git a/vvp/lexor.lex b/vvp/lexor.lex
index a2c19b7d5..0cf3fc272 100644
--- a/vvp/lexor.lex
+++ b/vvp/lexor.lex
@@ -183,6 +183,8 @@
 "%vpi_func/r" { return K_vpi_func_r; }
 "%disable"  { return K_disable; }
 "%fork"     { return K_fork; }
+"%alloc"    { return K_alloc; }
+"%free"     { return K_free; }
 
   /* Handle the specialized variable access functions. */
 
diff --git a/vvp/opcodes.txt b/vvp/opcodes.txt
index 7f7721fc8..be7f22f1c 100644
--- a/vvp/opcodes.txt
+++ b/vvp/opcodes.txt
@@ -49,6 +49,11 @@ This instruction adds the immediate value (no x or z bits) into the
 left vector. The imm value is limited to 16 significant bits, but it
 is zero extended to match any width.
 
+* %alloc <scope-label>
+
+This instruction allocates the storage for a new instance of an
+automatically allocated scope.
+
 * %and <bit-l>, <bit-r>, <wid>
 
 Perform the bitwise AND of the two vectors, and store the result in
@@ -362,6 +367,11 @@ The %fork instruction has no effect other than to push a child thread.
 
 See also %join.
 
+* %free <scope-label>
+
+This instruction de-allocates the storage for a previously allocated
+instance of as automatically allocated scope.
+
 
 * %inv <bit>, <wid>
 
diff --git a/vvp/parse.y b/vvp/parse.y
index 5bbcd714d..f404d1b7d 100644
--- a/vvp/parse.y
+++ b/vvp/parse.y
@@ -85,7 +85,7 @@ static struct __vpiModPath*modpath_dst = 0;
 %token K_THREAD K_TIMESCALE K_TRAN K_TRANIF0 K_TRANIF1 K_TRANVP K_UFUNC
 %token K_UDP K_UDP_C K_UDP_S
 %token K_VAR K_VAR_S K_VAR_I K_VAR_R K_vpi_call K_vpi_func K_vpi_func_r
-%token K_disable K_fork
+%token K_disable K_fork K_alloc K_free
 %token K_vpi_module K_vpi_time_precision K_file_names
 
 %token <text> T_INSTR
@@ -215,11 +215,11 @@ statement
      bits in the symbols list change. */
 
 	| T_LABEL K_UFUNC T_SYMBOL ',' T_NUMBER ',' symbols
-	  '(' symbols ')' symbol ';'
+	  '(' symbols ')' symbol symbol ';'
 		{ compile_ufunc($1, $3, $5,
 				$7.cnt, $7.vect,
 				$9.cnt, $9.vect,
-				$11); }
+				$11, $12); }
 
   /* Resolver statements are very much like functors. They are
      compiled to functors of a different mode. */
@@ -533,6 +533,11 @@ statement
 	| label_opt K_fork symbol ',' symbol ';'
 		{ compile_fork($1, $3, $5); }
 
+	| label_opt K_alloc symbol ';'
+		{ compile_alloc($1, $3); }
+
+	| label_opt K_free symbol ';'
+		{ compile_free($1, $3); }
 
   /* Scope statements come in two forms. There are the scope
      declaration and the scope recall. The declarations create the
diff --git a/vvp/ufunc.cc b/vvp/ufunc.cc
index 0d769cc0c..4570a0275 100644
--- a/vvp/ufunc.cc
+++ b/vvp/ufunc.cc
@@ -37,17 +37,19 @@
 
 ufunc_core::ufunc_core(unsigned owid, vvp_net_t*ptr,
 		       unsigned nports, vvp_net_t**ports,
-		       vvp_code_t sa, struct __vpiScope*run_scope,
-		       char*result_label)
+		       vvp_code_t sa, struct __vpiScope*call_scope,
+		       char*result_label, char*scope_label)
 : vvp_wide_fun_core(ptr, nports)
 {
       owid_ = owid;
       ports_ = ports;
       code_ = sa;
       thread_ = 0;
-      scope_ = run_scope;
+      call_scope_ = call_scope;
 
       functor_ref_lookup(&result_, result_label);
+
+      compile_vpi_lookup((vpiHandle*)(&func_scope_), scope_label);
 }
 
 ufunc_core::~ufunc_core()
@@ -55,7 +57,7 @@ ufunc_core::~ufunc_core()
 }
 
 /*
- * This method is called by the %fork_ufunc function to prepare the
+ * This method is called by the %exec_ufunc function to prepare the
  * input variables of the function for execution. The method copies
  * the input values collected by the core to the variables.
  */
@@ -72,7 +74,7 @@ void ufunc_core::assign_bits_to_ports(void)
 }
 
 /*
- * This method is called by the %join_ufunc instruction to copy the
+ * This method is called by the %exec_ufunc instruction to copy the
  * result from the return code variable and deliver it to the output
  * of the functor, back into the netlist.
  */
@@ -105,7 +107,7 @@ void ufunc_core::recv_real_from_inputs(unsigned port)
 void ufunc_core::invoke_thread_()
 {
       if (thread_ == 0) {
-	    thread_ = vthread_new(code_, scope_);
+	    thread_ = vthread_new(code_, call_scope_);
 	    schedule_vthread(thread_, 0);
       }
 }
@@ -123,15 +125,15 @@ void ufunc_core::invoke_thread_()
 void compile_ufunc(char*label, char*code, unsigned wid,
 		   unsigned argc,  struct symb_s*argv,
 		   unsigned portc, struct symb_s*portv,
-		   struct symb_s retv)
+		   struct symb_s retv, struct symb_s scope)
 {
 	/* The input argument list and port list must have the same
 	   sizes, since internally we will be mapping the inputs list
 	   to the ports list. */
       assert(argc == portc);
 
-      struct __vpiScope*run_scope = vpip_peek_current_scope();
-      assert(run_scope);
+      struct __vpiScope*call_scope = vpip_peek_current_scope();
+      assert(call_scope);
 
 	/* Construct some phantom code that is the thread of the
 	   function call. The first instruction, at the start_address
@@ -139,32 +141,19 @@ void compile_ufunc(char*label, char*code, unsigned wid,
 	   The last instruction is the usual %end. So the thread looks
 	   like this:
 
-	      %fork_ufunc <core>;
-	      %join;
-	      %join_ufunc;
+	      %exec_ufunc <core>;
 	      %end;
 
-	   The %fork_ufunc starts the user defined function by copying
-	   the input values into local regs, forking a thread and
-	   pushing that thread. The %join waits on that thread. The
-	   $join_ufunc then copies the output values to the
-	   destination net functors. */
+	   The %exec_ufunc copies the input values into local regs,
+           runs the function code, then copies the output values to
+           the destination net functors. */
 
       vvp_code_t start_code = codespace_allocate();
-      start_code->opcode = of_FORK_UFUNC;
+      start_code->opcode = of_EXEC_UFUNC;
       code_label_lookup(start_code, code);
 
-      { vvp_code_t codep = codespace_allocate();
-	codep->opcode = &of_JOIN;
-      }
-
-      vvp_code_t ujoin_code;
-      ujoin_code = codespace_allocate();
-      ujoin_code->opcode = &of_JOIN_UFUNC;
-
-      { vvp_code_t codep = codespace_allocate();
-	codep->opcode = &of_END;
-      }
+      vvp_code_t end_code = codespace_allocate();
+      end_code->opcode = &of_END;
 
 	/* Run through the function ports (which are related to but
 	   not the same as the input ports) and arrange for their
@@ -179,14 +168,13 @@ void compile_ufunc(char*label, char*code, unsigned wid,
 	   that will contain the execution. */
       vvp_net_t*ptr = new vvp_net_t;
       ufunc_core*fcore = new ufunc_core(wid, ptr, portc, ports,
-					start_code, run_scope,
-					retv.text);
+					start_code, call_scope,
+					retv.text, scope.text);
       ptr->fun = fcore;
       define_functor_symbol(label, ptr);
       free(label);
 
       start_code->ufunc_core_ptr = fcore;
-      ujoin_code->ufunc_core_ptr = fcore;
 
       wide_inputs_connect(fcore, argc, argv);
 
diff --git a/vvp/ufunc.h b/vvp/ufunc.h
index b76bddea4..28b88e9cd 100644
--- a/vvp/ufunc.h
+++ b/vvp/ufunc.h
@@ -53,11 +53,13 @@ class ufunc_core : public vvp_wide_fun_core {
       ufunc_core(unsigned ow, vvp_net_t*ptr,
 		 unsigned nports, vvp_net_t**ports,
 		 vvp_code_t start_address,
-		 struct __vpiScope*run_scope,
-		 char*result_label);
+		 struct __vpiScope*call_scope,
+		 char*result_label,
+		 char*scope_label);
       ~ufunc_core();
 
-      struct __vpiScope*scope() { return scope_; }
+      struct __vpiScope*call_scope() { return call_scope_; }
+      struct __vpiScope*func_scope() { return func_scope_; }
 
       void assign_bits_to_ports(void);
       void finish_thread(vthread_t thr);
@@ -80,7 +82,8 @@ class ufunc_core : public vvp_wide_fun_core {
 	// This is a thread to execute the behavioral portion of the
 	// function.
       vthread_t thread_;
-      struct __vpiScope*scope_;
+      struct __vpiScope*call_scope_;
+      struct __vpiScope*func_scope_;
       vvp_code_t code_;
 
 	// Where the result will be.
diff --git a/vvp/vpi_priv.h b/vvp/vpi_priv.h
index b857eb0e0..e2b0634a7 100644
--- a/vvp/vpi_priv.h
+++ b/vvp/vpi_priv.h
@@ -173,12 +173,18 @@ struct __vpiScope {
       unsigned lineno;
       unsigned def_file_idx;
       unsigned def_lineno;
+      bool is_automatic;
 	/* The scope has a system time of its own. */
       struct __vpiSystemTime scoped_time;
       struct __vpiSystemTime scoped_realtime;
 	/* Keep an array of internal scope items. */
       struct __vpiHandle**intern;
       unsigned nintern;
+        /* Keep an array of items to be automatically allocated */
+      struct automatic_hooks_s**item;
+      unsigned nitem;
+        /* Keep a list of freed contexts. */
+      vvp_context_t free_context;
 	/* Keep a list of threads in the scope. */
       vthread_t threads;
       signed int time_units :8;
@@ -187,6 +193,7 @@ struct __vpiScope {
 
 extern struct __vpiScope* vpip_peek_current_scope(void);
 extern void vpip_attach_to_current_scope(vpiHandle obj);
+extern void vpip_add_item_to_current_scope(automatic_hooks_s*item);
 extern vpiHandle vpip_make_root_iterator(void);
 extern void vpip_make_root_iterator(struct __vpiHandle**&table,
 				    unsigned&ntable);
diff --git a/vvp/vpi_scope.cc b/vvp/vpi_scope.cc
index 54875655d..1a18e0f56 100644
--- a/vvp/vpi_scope.cc
+++ b/vvp/vpi_scope.cc
@@ -316,6 +316,26 @@ static void attach_to_scope_(struct __vpiScope*scope, vpiHandle obj)
       scope->intern[idx] = obj;
 }
 
+static void add_item_to_scope_(struct __vpiScope*scope, automatic_hooks_s*item)
+{
+      assert(scope);
+
+        // there is no need to record items for static scopes
+      if (!scope->is_automatic) return;
+
+      unsigned idx = scope->nitem++;
+      item->context_idx = 1 + idx;
+
+      if (scope->item == 0)
+	    scope->item = (automatic_hooks_s**)
+		  malloc(sizeof(automatic_hooks_s*));
+      else
+	    scope->item = (automatic_hooks_s**)
+		  realloc(scope->item, sizeof(automatic_hooks_s*)*scope->nitem);
+
+      scope->item[idx] = item;
+}
+
 /*
  * When the compiler encounters a scope declaration, this function
  * creates and initializes a __vpiScope object with the requested name
@@ -330,20 +350,33 @@ compile_scope_decl(char*label, char*type, char*name, const char*tname,
       struct __vpiScope*scope = new struct __vpiScope;
       count_vpi_scopes += 1;
 
-      if (strcmp(type,"module") == 0)
+      if (strcmp(type,"module") == 0) {
 	    scope->base.vpi_type = &vpip_scope_module_rt;
-      else if (strcmp(type,"function") == 0)
+	    scope->is_automatic = false;
+      } else if (strcmp(type,"autofunction") == 0) {
 	    scope->base.vpi_type = &vpip_scope_function_rt;
-      else if (strcmp(type,"task") == 0)
+	    scope->is_automatic = true;
+      } else if (strcmp(type,"function") == 0) {
+	    scope->base.vpi_type = &vpip_scope_function_rt;
+	    scope->is_automatic = false;
+      } else if (strcmp(type,"autotask") == 0) {
 	    scope->base.vpi_type = &vpip_scope_task_rt;
-      else if (strcmp(type,"fork") == 0)
+	    scope->is_automatic = true;
+      } else if (strcmp(type,"task") == 0) {
+	    scope->base.vpi_type = &vpip_scope_task_rt;
+	    scope->is_automatic = false;
+      } else if (strcmp(type,"fork") == 0) {
 	    scope->base.vpi_type = &vpip_scope_fork_rt;
-      else if (strcmp(type,"begin") == 0)
+	    scope->is_automatic = false;
+      } else if (strcmp(type,"begin") == 0) {
 	    scope->base.vpi_type = &vpip_scope_begin_rt;
-      else if (strcmp(type,"generate") == 0)
+	    scope->is_automatic = false;
+      } else if (strcmp(type,"generate") == 0) {
 	    scope->base.vpi_type = &vpip_scope_begin_rt;
-      else {
+	    scope->is_automatic = false;
+      } else {
 	    scope->base.vpi_type = &vpip_scope_module_rt;
+	    scope->is_automatic = false;
 	    assert(0);
       }
 
@@ -357,6 +390,9 @@ compile_scope_decl(char*label, char*type, char*name, const char*tname,
       scope->def_lineno  = (unsigned) def_lineno;
       scope->intern = 0;
       scope->nintern = 0;
+      scope->item = 0;
+      scope->nitem = 0;
+      scope->free_context = 0;
       scope->threads = 0;
 
       current_scope = scope;
@@ -421,3 +457,8 @@ void vpip_attach_to_current_scope(vpiHandle obj)
 {
       attach_to_scope_(current_scope, obj);
 }
+
+void vpip_add_item_to_current_scope(automatic_hooks_s*item)
+{
+      add_item_to_scope_(current_scope, item);
+}
diff --git a/vvp/vthread.cc b/vvp/vthread.cc
index 9b2e56550..3bb7bd353 100644
--- a/vvp/vthread.cc
+++ b/vvp/vthread.cc
@@ -43,7 +43,7 @@
 # define TOP_BIT (1UL << (CPU_WORD_BITS-1))
 
 /*
- * This vhtread_s structure describes all there is to know about a
+ * This vthread_s structure describes all there is to know about a
  * thread, including its program counter, all the private bits it
  * holds, and its place in other lists.
  *
@@ -117,12 +117,15 @@ struct vthread_s {
       struct vthread_s*wait_next;
 	/* These are used to keep the thread in a scope. */
       struct vthread_s*scope_next, *scope_prev;
-
+	/* These are used to access automatically allocated items. */
+      vvp_context_t wt_context, rd_context;
 	/* These are used to pass non-blocking event control information. */
       vvp_net_t*event;
       uint64_t ecount;
 };
 
+struct vthread_s*running_thread = 0;
+
 // this table maps the thread special index bit addresses to
 // vvp_bit4_t bit values.
 static vvp_bit4_t thr_index_to_bit4[4] = { BIT4_0, BIT4_1, BIT4_X, BIT4_Z };
@@ -298,6 +301,43 @@ static void multiply_array_imm(unsigned long*res, unsigned long*val,
       }
 }
 
+/*
+ * Allocate a context for use by a child thread. By preference, use
+ * the last freed context. If none available, create a new one.
+ */
+static vvp_context_t vthread_alloc_context(__vpiScope*scope)
+{
+      assert(scope->is_automatic);
+
+      vvp_context_t context = scope->free_context;
+      if (context) {
+            scope->free_context = vvp_get_next_context(context);
+            for (unsigned idx = 0 ; idx < scope->nitem ; idx += 1) {
+                  scope->item[idx]->reset_instance(context);
+            }
+      } else {
+            context = vvp_allocate_context(scope->nitem);
+            for (unsigned idx = 0 ; idx < scope->nitem ; idx += 1) {
+                  scope->item[idx]->alloc_instance(context);
+            }
+      }
+
+      return context;
+}
+
+/*
+ * Free a context previously allocated to a child thread by pushing it
+ * onto the freed context stack.
+ */
+static void vthread_free_context(vvp_context_t context, __vpiScope*scope)
+{
+      assert(scope->is_automatic);
+      assert(context);
+
+      vvp_set_next_context(context, scope->free_context);
+      scope->free_context = context;
+}
+
 /*
  * Create a new thread with the given start address.
  */
@@ -309,6 +349,8 @@ vthread_t vthread_new(vvp_code_t pc, struct __vpiScope*scope)
       thr->child  = 0;
       thr->parent = 0;
       thr->wait_next = 0;
+      thr->wt_context = 0;
+      thr->rd_context = 0;
 
 	/* If the target scope never held a thread, then create a
 	   header cell for it. This is a stub to make circular lists
@@ -334,7 +376,6 @@ vthread_t vthread_new(vvp_code_t pc, struct __vpiScope*scope)
       thr->is_scheduled = 0;
       thr->i_have_ended = 0;
       thr->waiting_for_event = 0;
-      thr->is_scheduled = 0;
       thr->fork_count   = 0;
       thr->event  = 0;
       thr->ecount = 0;
@@ -409,6 +450,8 @@ void vthread_run(vthread_t thr)
 	    assert(thr->is_scheduled);
 	    thr->is_scheduled = 0;
 
+            running_thread = thr;
+
 	    for (;;) {
 		  vvp_code_t cp = thr->pc;
 		  thr->pc += 1;
@@ -423,6 +466,7 @@ void vthread_run(vthread_t thr)
 
 	    thr = tmp;
       }
+      running_thread = 0;
 }
 
 /*
@@ -486,6 +530,18 @@ void vthread_schedule_list(vthread_t thr)
       schedule_vthread(thr, 0);
 }
 
+vvp_context_item_t vthread_get_wt_context_item(unsigned context_idx)
+{
+      assert(running_thread && running_thread->wt_context);
+      return vvp_get_context_item(running_thread->wt_context, context_idx);
+}
+
+vvp_context_item_t vthread_get_rd_context_item(unsigned context_idx)
+{
+      assert(running_thread && running_thread->rd_context);
+      return vvp_get_context_item(running_thread->rd_context, context_idx);
+}
+
 bool of_ABS_WR(vthread_t thr, vvp_code_t cp)
 {
       unsigned dst = cp->bit_idx[0];
@@ -495,6 +551,18 @@ bool of_ABS_WR(vthread_t thr, vvp_code_t cp)
       return true;
 }
 
+bool of_ALLOC(vthread_t thr, vvp_code_t cp)
+{
+        /* Allocate a context. */
+      vvp_context_t child_context = vthread_alloc_context(cp->scope);
+
+        /* Push the allocated context onto the write context stack. */
+      vvp_set_next_context(child_context, thr->wt_context);
+      thr->wt_context = child_context;
+
+      return true;
+}
+
 static bool of_AND_wide(vthread_t thr, vvp_code_t cp)
 {
       unsigned idx1 = cp->bit_idx[0];
@@ -2308,6 +2376,12 @@ bool of_FORCE_X0(vthread_t thr, vvp_code_t cp)
 bool of_FORK(vthread_t thr, vvp_code_t cp)
 {
       vthread_t child = vthread_new(cp->cptr2, cp->scope);
+      if (cp->scope->is_automatic) {
+              /* The context allocated for this child is the top entry
+                 on the write context stack. */
+            child->wt_context = thr->wt_context;
+            child->rd_context = thr->wt_context;
+      }
 
       child->child  = thr->child;
       child->parent = thr;
@@ -2324,9 +2398,23 @@ bool of_FORK(vthread_t thr, vvp_code_t cp)
       if (cp->scope->base.vpi_type->type_code == vpiFunction) {
 	    child->is_scheduled = 1;
 	    vthread_run(child);
+            running_thread = thr;
       } else {
 	    schedule_vthread(child, 0, true);
       }
+
+      return true;
+}
+
+bool of_FREE(vthread_t thr, vvp_code_t cp)
+{
+        /* Pop the child context from the read context stack. */
+      vvp_context_t child_context = thr->rd_context;
+      thr->rd_context = vvp_get_next_context(child_context);
+
+        /* Free the context. */
+      vthread_free_context(child_context, cp->scope);
+
       return true;
 }
 
@@ -2610,6 +2698,15 @@ bool of_JOIN(vthread_t thr, vvp_code_t cp)
 
       assert(thr->fork_count > 0);
 
+      if (thr->wt_context != thr->rd_context) {
+              /* Pop the child context from the write context stack. */
+            vvp_context_t child_context = thr->wt_context;
+            thr->wt_context = vvp_get_next_context(child_context);
+
+              /* Push the child context onto the read context stack */
+            vvp_set_next_context(child_context, thr->rd_context);
+            thr->rd_context = child_context;
+      }
 
 	/* If the child has already ended, reap it now. */
       if (thr->child->i_have_ended) {
@@ -4080,7 +4177,6 @@ bool of_SUBI(vthread_t thr, vvp_code_t cp)
 
 bool of_VPI_CALL(vthread_t thr, vvp_code_t cp)
 {
-	// printf("thread %p: %%vpi_call\n", thr);
       vpip_execute_vpi_call(thr, cp->handle);
 
       if (schedule_stopped()) {
@@ -4104,13 +4200,19 @@ bool of_WAIT(vthread_t thr, vvp_code_t cp)
       assert(! thr->waiting_for_event);
       thr->waiting_for_event = 1;
 
-      vvp_net_t*net = cp->net;
-	/* Get the functor as a waitable_hooks_s object. */
-      waitable_hooks_s*ep = dynamic_cast<waitable_hooks_s*> (net->fun);
-      assert(ep);
 	/* Add this thread to the list in the event. */
-      thr->wait_next = ep->threads;
-      ep->threads = thr;
+      vvp_net_fun_t*fun = cp->net->fun;
+      if (fun->context_idx) {
+            waitable_state_s*es = static_cast<waitable_state_s*>
+                  (vthread_get_wt_context_item(fun->context_idx));
+            thr->wait_next = es->threads;
+            es->threads = thr;
+      } else {
+            waitable_hooks_s*ep = dynamic_cast<waitable_hooks_s*> (fun);
+            assert(ep);
+            thr->wait_next = ep->threads;
+            ep->threads = thr;
+      }
 	/* Return false to suspend this thread. */
       return false;
 }
@@ -4185,46 +4287,57 @@ bool of_ZOMBIE(vthread_t thr, vvp_code_t)
 }
 
 /*
- * These are phantom opcode used to call user defined functions.
- * They are used in code generated by the .ufunc statement. They
- * contain a pointer to executable code of the function, and to a
- * ufunc_core object that has all the port information about the
+ * This is a phantom opcode used to call user defined functions. It
+ * is used in code generated by the .ufunc statement. It contains a
+ * pointer to the executable code of the function and a pointer to
+ * a ufunc_core object that has all the port information about the
  * function.
  */
-bool of_FORK_UFUNC(vthread_t thr, vvp_code_t cp)
+bool of_EXEC_UFUNC(vthread_t thr, vvp_code_t cp)
 {
+      struct __vpiScope*child_scope = cp->ufunc_core_ptr->func_scope();
+      assert(child_scope);
+
+      assert(thr->child == 0);
+      assert(thr->fork_count == 0);
+
+        /* We can take a number of shortcuts because we know that a
+           continuous assignment can only occur in a static scope. */
+      assert(thr->wt_context == 0);
+      assert(thr->rd_context == 0);
+
+        /* If an automatic function, allocate a context for this call. */
+      vvp_context_t child_context = 0;
+      if (child_scope->is_automatic) {
+            child_context = vthread_alloc_context(child_scope);
+            thr->wt_context = child_context;
+            thr->rd_context = child_context;
+      }
 	/* Copy all the inputs to the ufunc object to the port
 	   variables of the function. This copies all the values
 	   atomically. */
       cp->ufunc_core_ptr->assign_bits_to_ports();
 
-      assert(thr->child == 0);
-      assert(thr->fork_count == 0);
+	/* Create a temporary thread and run it immediately. A function
+           may not contain any blocking statements, so vthread_run() can
+           only return when the %end opcode is reached. */
+      vthread_t child = vthread_new(cp->cptr, child_scope);
+      child->wt_context = child_context;
+      child->rd_context = child_context;
+      child->is_scheduled = 1;
+      vthread_run(child);
+      running_thread = thr;
 
-	/* Create a temporary thread, and push its execution. This is
-	   done so that the assign_bits_to_ports above is atomic with
-	   this startup. */
-      vthread_t child = vthread_new(cp->cptr, cp->ufunc_core_ptr->scope());
-
-      child->child = 0;
-      child->parent = thr;
-      thr->child = child;
-
-      thr->fork_count += 1;
-      schedule_vthread(child, 0, true);
-
-	/* After this function, the .ufunc code has placed an of_JOIN
-	   to pause this thread. Since the child was pushed by the
-	   flag to schedule_vthread, the called function starts up
-	   immediately. */
-      return true;
-}
-
-bool of_JOIN_UFUNC(vthread_t thr, vvp_code_t cp)
-{
 	/* Now copy the output from the result variable to the output
 	   ports of the .ufunc device. */
       cp->ufunc_core_ptr->finish_thread(thr);
 
+        /* If an automatic function, free the context for this call. */
+      if (child_scope->is_automatic) {
+            vthread_free_context(child_context, child_scope);
+            thr->wt_context = 0;
+            thr->rd_context = 0;
+      }
+
       return true;
 }
diff --git a/vvp/vthread.h b/vvp/vthread.h
index b6ee59cf0..f578b4ef8 100644
--- a/vvp/vthread.h
+++ b/vvp/vthread.h
@@ -62,6 +62,29 @@ extern void vthread_run(vthread_t thr);
  */
 extern void vthread_schedule_list(vthread_t thr);
 
+/*
+ * This function returns a handle to an item in the writable context
+ * of the currently running thread. Normally the writable context is
+ * the context allocated to the scope associated with that thread.
+ * However, between executing a %alloc instruction and executing the
+ * associated %fork instruction, the writable context changes to the
+ * newly allocated context, thus allowing the input parameters of an
+ * automatic task or function to be written to the task/function local
+ * variables.
+ */
+extern vvp_context_item_t vthread_get_wt_context_item(unsigned context_idx);
+
+/*
+ * This function returns a handle to an item in the readable context
+ * of the currently running thread. Normally the readable context is
+ * the context allocated to the scope associated with that thread.
+ * However, between executing a %join instruction and executing the
+ * associated %free instruction, the readable context changes to the
+ * context allocated to the newly joined thread, thus allowing the
+ * output parameters of an automatic task or function to be read from
+ * the task/function local variables.
+ */
+extern vvp_context_item_t vthread_get_rd_context_item(unsigned context_idx);
 
 /*
  * Return a bit from the thread's bit space. These are used, for
diff --git a/vvp/vvp_net.cc b/vvp/vvp_net.cc
index 1a8ab4c86..9c8681e2a 100644
--- a/vvp/vvp_net.cc
+++ b/vvp/vvp_net.cc
@@ -990,6 +990,20 @@ void vvp_vector4_t::change_z2x()
       }
 }
 
+void vvp_vector4_t::set_to_x()
+{
+      if (size_ <= BITS_PER_WORD) {
+	    abits_val_ = vvp_vector4_t::WORD_X_ABITS;
+            bbits_val_ = vvp_vector4_t::WORD_X_BBITS;
+      } else {
+	    unsigned words = (size_+BITS_PER_WORD-1) / BITS_PER_WORD;
+	    for (unsigned idx = 0 ;  idx < words ;  idx += 1) {
+		  abits_ptr_[idx] = vvp_vector4_t::WORD_X_ABITS;
+                  bbits_ptr_[idx] = vvp_vector4_t::WORD_X_BBITS;
+            }
+      }
+}
+
 char* vvp_vector4_t::as_string(char*buf, size_t buf_len)
 {
       char*res = buf;
@@ -1284,9 +1298,12 @@ bool vector4_to_value(const vvp_vector4_t&vec, double&val, bool signed_flag)
       return flag;
 }
 
-vvp_vector4array_t::vvp_vector4array_t(unsigned width, unsigned words)
-: width_(width), words_(words)
+vvp_vector4array_t::vvp_vector4array_t(unsigned width, unsigned words,
+                                       bool is_automatic)
+: width_(width), words_(words), array_(0)
 {
+      if (is_automatic) return;
+
       array_ = new v4cell[words_];
 
       if (width_ <= vvp_vector4_t::BITS_PER_WORD) {
@@ -1314,30 +1331,79 @@ vvp_vector4array_t::~vvp_vector4array_t()
       }
 }
 
+void vvp_vector4array_t::alloc_instance(vvp_context_t context)
+{
+      v4cell*array = new v4cell[words_];
+
+      if (width_ <= vvp_vector4_t::BITS_PER_WORD) {
+	    for (unsigned idx = 0 ; idx < words_ ; idx += 1) {
+		  array[idx].abits_val_ = vvp_vector4_t::WORD_X_ABITS;
+		  array[idx].bbits_val_ = vvp_vector4_t::WORD_X_BBITS;
+	    }
+      } else {
+	    for (unsigned idx = 0 ; idx < words_ ; idx += 1) {
+		  array[idx].abits_ptr_ = 0;
+		  array[idx].bbits_ptr_ = 0;
+	    }
+      }
+
+      vvp_set_context_item(context, context_idx, array);
+}
+
+void vvp_vector4array_t::reset_instance(vvp_context_t context)
+{
+      v4cell*cell = static_cast<v4cell*>
+            (vvp_get_context_item(context, context_idx));
+
+      if (width_ <= vvp_vector4_t::BITS_PER_WORD) {
+	    for (unsigned idx = 0 ; idx < words_ ; idx += 1) {
+		  cell->abits_val_ = vvp_vector4_t::WORD_X_ABITS;
+		  cell->bbits_val_ = vvp_vector4_t::WORD_X_BBITS;
+                  cell++;
+	    }
+      } else {
+            unsigned cnt = (width_ + vvp_vector4_t::BITS_PER_WORD-1)/vvp_vector4_t::BITS_PER_WORD;
+	    for (unsigned idx = 0 ; idx < words_ ; idx += 1) {
+		  if (cell->abits_ptr_) {
+                        for (unsigned n = 0 ; n < cnt ; n += 1) {
+                  	    cell->abits_ptr_[n] = vvp_vector4_t::WORD_X_ABITS;
+                  	    cell->bbits_ptr_[n] = vvp_vector4_t::WORD_X_BBITS;
+                        }
+                  }
+                  cell++;
+	    }
+      }
+}
+
 void vvp_vector4array_t::set_word(unsigned index, const vvp_vector4_t&that)
 {
       assert(index < words_);
       assert(that.size_ == width_);
 
-      v4cell&cell = array_[index];
+      v4cell*cell;
+      if (context_idx)
+            cell = static_cast<v4cell*>
+               (vthread_get_wt_context_item(context_idx)) + index;
+      else
+            cell = &(array_[index]);
 
       if (width_ <= vvp_vector4_t::BITS_PER_WORD) {
-	    cell.abits_val_ = that.abits_val_;
-	    cell.bbits_val_ = that.bbits_val_;
+	    cell->abits_val_ = that.abits_val_;
+	    cell->bbits_val_ = that.bbits_val_;
 	    return;
       }
 
       unsigned cnt = (width_ + vvp_vector4_t::BITS_PER_WORD-1)/vvp_vector4_t::BITS_PER_WORD;
 
-      if (cell.abits_ptr_ == 0) {
-	    cell.abits_ptr_ = new unsigned long[2*cnt];
-	    cell.bbits_ptr_ = cell.abits_ptr_ + cnt;
+      if (cell->abits_ptr_ == 0) {
+	    cell->abits_ptr_ = new unsigned long[2*cnt];
+	    cell->bbits_ptr_ = cell->abits_ptr_ + cnt;
       }
 
       for (unsigned idx = 0 ; idx < cnt ; idx += 1)
-	    cell.abits_ptr_[idx] = that.abits_ptr_[idx];
+	    cell->abits_ptr_[idx] = that.abits_ptr_[idx];
       for (unsigned idx = 0 ; idx < cnt ; idx += 1)
-	    cell.bbits_ptr_[idx] = that.bbits_ptr_[idx];
+	    cell->bbits_ptr_[idx] = that.bbits_ptr_[idx];
 }
 
 vvp_vector4_t vvp_vector4array_t::get_word(unsigned index) const
@@ -1347,26 +1413,31 @@ vvp_vector4_t vvp_vector4array_t::get_word(unsigned index) const
 
       assert(index < words_);
 
-      v4cell&cell = array_[index];
+      v4cell*cell;
+      if (context_idx)
+            cell = static_cast<v4cell*>
+               (vthread_get_rd_context_item(context_idx)) + index;
+      else
+            cell = &(array_[index]);
 
       if (width_ <= vvp_vector4_t::BITS_PER_WORD) {
 	    vvp_vector4_t res;
 	    res.size_ = width_;
-	    res.abits_val_ = cell.abits_val_;
-	    res.bbits_val_ = cell.bbits_val_;
+	    res.abits_val_ = cell->abits_val_;
+	    res.bbits_val_ = cell->bbits_val_;
 	    return res;
       }
 
       vvp_vector4_t res (width_, BIT4_X);
-      if (cell.abits_ptr_ == 0)
+      if (cell->abits_ptr_ == 0)
 	    return res;
 
       unsigned cnt = (width_ + vvp_vector4_t::BITS_PER_WORD-1)/vvp_vector4_t::BITS_PER_WORD;
 
       for (unsigned idx = 0 ; idx < cnt ; idx += 1)
-	    res.abits_ptr_[idx] = cell.abits_ptr_[idx];
+	    res.abits_ptr_[idx] = cell->abits_ptr_[idx];
       for (unsigned idx = 0 ; idx < cnt ; idx += 1)
-	    res.bbits_ptr_[idx] = cell.bbits_ptr_[idx];
+	    res.bbits_ptr_[idx] = cell->bbits_ptr_[idx];
 
       return res;
 
@@ -2350,6 +2421,21 @@ vvp_fun_signal::vvp_fun_signal(unsigned wid, vvp_bit4_t init)
 {
 }
 
+void vvp_fun_signal::alloc_instance(vvp_context_t context)
+{
+      unsigned wid = bits4_.size();
+
+      vvp_set_context_item(context, context_idx, new vvp_vector4_t(wid));
+}
+
+void vvp_fun_signal::reset_instance(vvp_context_t context)
+{
+      vvp_vector4_t*bits = static_cast<vvp_vector4_t*>
+            (vvp_get_context_item(context, context_idx));
+
+      bits->set_to_x();
+}
+
 /*
  * Nets simply reflect their input to their output.
  *
@@ -2361,6 +2447,10 @@ vvp_fun_signal::vvp_fun_signal(unsigned wid, vvp_bit4_t init)
  * herein is to keep a "needs_init_" flag that is turned false after
  * the first propagation, and forces the first propagation to happen
  * even if it matches the initial value.
+ *
+ * Continuous and forced assignments are not permitted on automatic
+ * variables. So we only need incur the overhead of checking for an
+ * automatic variable when we are doing a normal unmasked assign.
  */
 void vvp_fun_signal::recv_vec4(vvp_net_ptr_t ptr, const vvp_vector4_t&bit)
 {
@@ -2370,8 +2460,15 @@ void vvp_fun_signal::recv_vec4(vvp_net_ptr_t ptr, const vvp_vector4_t&bit)
 		 copy the bits, otherwise we need to see if there are
 		 any holes in the mask so we can set those bits. */
 	    if (assign_mask_.size() == 0) {
-		  if (needs_init_ || !bits4_.eeq(bit)) {
-			bits4_ = bit;
+                  vvp_vector4_t*bits4;
+                  if (context_idx) {
+                        bits4 = static_cast<vvp_vector4_t*>
+                              (vthread_get_wt_context_item(context_idx));
+                  } else {
+                        bits4 = &bits4_;
+                  }
+		  if (needs_init_ || !bits4->eeq(bit)) {
+			*bits4 = bit;
 			needs_init_ = false;
 			calculate_output_(ptr);
 		  }
@@ -2426,9 +2523,16 @@ void vvp_fun_signal::recv_vec4_pv(vvp_net_ptr_t ptr, const vvp_vector4_t&bit,
       switch (ptr.port()) {
 	  case 0: // Normal input
 	    if (assign_mask_.size() == 0) {
+                  vvp_vector4_t*bits4;
+                  if (context_idx) {
+                        bits4 = static_cast<vvp_vector4_t*>
+                              (vthread_get_wt_context_item(context_idx));
+                  } else {
+                        bits4 = &bits4_;
+                  }
 		  for (unsigned idx = 0 ;  idx < wid ;  idx += 1) {
-			if (base+idx >= bits4_.size()) break;
-			bits4_.set_bit(base+idx, bit.value(idx));
+			if (base+idx >= bits4->size()) break;
+			bits4->set_bit(base+idx, bit.value(idx));
 		  }
 		  needs_init_ = false;
 		  calculate_output_(ptr);
@@ -2499,9 +2603,12 @@ void vvp_fun_signal::calculate_output_(vvp_net_ptr_t ptr)
 			bits.set_bit(idx, force_.value(idx));
 	    }
 	    vvp_send_vec4(ptr.ptr()->out, bits);
-
+      } else if (context_idx) {
+            vvp_vector4_t*bits4 = static_cast<vvp_vector4_t*>
+                  (vthread_get_wt_context_item(context_idx));
+            vvp_send_vec4(ptr.ptr()->out, *bits4);
       } else {
-	    vvp_send_vec4(ptr.ptr()->out, bits4_);
+            vvp_send_vec4(ptr.ptr()->out, bits4_);
       }
 
       run_vpi_callbacks();
@@ -2548,18 +2655,28 @@ unsigned vvp_fun_signal::size() const
 
 vvp_bit4_t vvp_fun_signal::value(unsigned idx) const
 {
-      if (force_mask_.size() && force_mask_.value(idx))
+      if (force_mask_.size() && force_mask_.value(idx)) {
 	    return force_.value(idx);
-      else
-	    return bits4_.value(idx);
+      } else if (context_idx) {
+            vvp_vector4_t*bits4 = static_cast<vvp_vector4_t*>
+                  (vthread_get_rd_context_item(context_idx));
+            return bits4->value(idx);
+      } else {
+            return bits4_.value(idx);
+      }
 }
 
 vvp_scalar_t vvp_fun_signal::scalar_value(unsigned idx) const
 {
-      if (force_mask_.size() && force_mask_.value(idx))
+      if (force_mask_.size() && force_mask_.value(idx)) {
 	    return vvp_scalar_t(force_.value(idx), 6, 6);
-      else
-	    return vvp_scalar_t(bits4_.value(idx), 6, 6);
+      } else if (context_idx) {
+            vvp_vector4_t*bits4 = static_cast<vvp_vector4_t*>
+                  (vthread_get_rd_context_item(context_idx));
+            return vvp_scalar_t(bits4->value(idx), 6, 6);
+      } else {
+            return vvp_scalar_t(bits4_.value(idx), 6, 6);
+      }
 }
 
 vvp_vector4_t vvp_fun_signal::vec4_value() const
@@ -2573,9 +2690,12 @@ vvp_vector4_t vvp_fun_signal::vec4_value() const
 			bits.set_bit(idx, force_.value(idx));
 	    }
 	    return bits;
-
+      } else if (context_idx) {
+            vvp_vector4_t*bits4 = static_cast<vvp_vector4_t*>
+                  (vthread_get_rd_context_item(context_idx));
+            return *bits4;
       } else {
-	    return bits4_;
+            return bits4_;
       }
 }
 
@@ -2764,12 +2884,30 @@ vvp_fun_signal_real::vvp_fun_signal_real()
       bits_ = 0.0;
 }
 
+void vvp_fun_signal_real::alloc_instance(vvp_context_t context)
+{
+      vvp_set_context_item(context, context_idx, new double);
+}
+
+void vvp_fun_signal_real::reset_instance(vvp_context_t context)
+{
+      double*bits = static_cast<double*>
+            (vvp_get_context_item(context, context_idx));
+
+      *bits = 0.0;
+}
+
 double vvp_fun_signal_real::real_value() const
 {
-      if (force_mask_.size())
+      if (force_mask_.size()) {
 	    return force_;
-      else
+      } else if (context_idx) {
+            double*bits = static_cast<double*>
+                  (vthread_get_rd_context_item(context_idx));
+            return *bits;
+      } else {
 	    return bits_;
+      }
 }
 
 /*
@@ -2787,8 +2925,15 @@ void vvp_fun_signal_real::recv_real(vvp_net_ptr_t ptr, double bit)
       switch (ptr.port()) {
 	  case 0:
 	    if (!continuous_assign_active_) {
-		  if (needs_init_ || !bits_equal(bits_,bit)) {
-			bits_ = bit;
+                  double*bits;
+                  if (context_idx) {
+                        bits = static_cast<double*>
+                              (vthread_get_wt_context_item(context_idx));
+                  } else {
+                        bits = &bits_;
+                  }
+		  if (needs_init_ || !bits_equal(*bits,bit)) {
+			*bits = bit;
 			needs_init_ = false;
 			vvp_send_real(ptr.ptr()->out, bit);
 			run_vpi_callbacks();
diff --git a/vvp/vvp_net.h b/vvp/vvp_net.h
index 97070e65e..d5d40fbe4 100644
--- a/vvp/vvp_net.h
+++ b/vvp/vvp_net.h
@@ -22,6 +22,7 @@
 # include  "config.h"
 # include  "vpi_user.h"
 # include  <stddef.h>
+# include  <stdlib.h>
 # include  <string.h>
 # include  <new>
 # include  <assert.h>
@@ -49,6 +50,58 @@ class  vvp_fun_part;
 
 class  vvp_delay_t;
 
+/*
+ * Storage for items declared in automatically allocated scopes (i.e. automatic
+ * tasks and functions).
+ */
+typedef void**vvp_context_t;
+
+typedef void*vvp_context_item_t;
+
+inline vvp_context_t vvp_allocate_context(unsigned nitem)
+{
+      return (vvp_context_t)malloc((1 + nitem) * sizeof(void*));
+}
+
+inline vvp_context_t vvp_get_next_context(vvp_context_t context)
+{
+      return (vvp_context_t)context[0];
+}
+
+inline void vvp_set_next_context(vvp_context_t context, vvp_context_t next)
+{
+      context[0] = next;
+}
+
+inline vvp_context_item_t vvp_get_context_item(vvp_context_t context,
+                                               unsigned item_idx)
+{
+      return (vvp_context_item_t)context[item_idx];
+}
+
+inline void vvp_set_context_item(vvp_context_t context, unsigned item_idx,
+                                 vvp_context_item_t item)
+{
+      context[item_idx] = item;
+}
+
+/*
+ * An "automatic" functor is one which may be associated with an automatically
+ * allocated scope item. This provides the infrastructure needed to allocate
+ * and access the state information for individual instances of the item. A
+ * context_idx value of 0 indicates a statically allocated item.
+ */
+struct automatic_hooks_s {
+
+      automatic_hooks_s() : context_idx(0) {}
+      virtual ~automatic_hooks_s() {}
+
+      virtual void alloc_instance(vvp_context_t context) {}
+      virtual void reset_instance(vvp_context_t context) {}
+
+      unsigned context_idx;
+};
+
 /*
  * This is the set of Verilog 4-value bit values. Scalars have this
  * value along with strength, vectors are a collection of these
@@ -169,6 +222,9 @@ class vvp_vector4_t {
 	// Change all Z bits to X bits.
       void change_z2x();
 
+	// Change all bits to X bits.
+      void set_to_x();
+
 	// Display the value into the buf as a string.
       char*as_string(char*buf, size_t buf_len);
 
@@ -402,12 +458,15 @@ extern bool vector4_to_value(const vvp_vector4_t&a, double&val, bool is_signed);
 /*
  * vvp_vector4array_t
  */
-class vvp_vector4array_t {
+class vvp_vector4array_t : public automatic_hooks_s {
 
     public:
-      vvp_vector4array_t(unsigned width, unsigned words);
+      vvp_vector4array_t(unsigned width, unsigned words, bool is_automatic);
       ~vvp_vector4array_t();
 
+      void alloc_instance(vvp_context_t context);
+      void reset_instance(vvp_context_t context);
+
       unsigned width() const { return width_; }
       unsigned words() const { return words_; }
 
@@ -866,7 +925,7 @@ struct vvp_net_t {
  * operand to a vvp_vector4_t and pass it on to the recv_vec4 or
  * recv_vec4_pv method.
  */
-class vvp_net_fun_t {
+class vvp_net_fun_t : public automatic_hooks_s {
 
     public:
       vvp_net_fun_t();
@@ -1120,6 +1179,9 @@ class vvp_fun_signal  : public vvp_fun_signal_vec {
     public:
       explicit vvp_fun_signal(unsigned wid, vvp_bit4_t init=BIT4_X);
 
+      void alloc_instance(vvp_context_t context);
+      void reset_instance(vvp_context_t context);
+
       void recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit);
       void recv_vec8(vvp_net_ptr_t port, const vvp_vector8_t&bit);
 
@@ -1188,6 +1250,9 @@ class vvp_fun_signal_real  : public vvp_fun_signal_base {
     public:
       explicit vvp_fun_signal_real();
 
+      void alloc_instance(vvp_context_t context);
+      void reset_instance(vvp_context_t context);
+
 	//void recv_vec4(vvp_net_ptr_t port, const vvp_vector4_t&bit);
       void recv_real(vvp_net_ptr_t port, double bit);
 
diff --git a/vvp/words.cc b/vvp/words.cc
index b73574871..fb048d697 100644
--- a/vvp/words.cc
+++ b/vvp/words.cc
@@ -37,6 +37,7 @@ static void __compile_var_real(char*label, char*name,
       vvp_fun_signal_real*fun = new vvp_fun_signal_real;
       vvp_net_t*net = new vvp_net_t;
       net->fun = fun;
+      vpip_add_item_to_current_scope(fun);
       define_functor_symbol(label, net);
 
       vpiHandle obj = vpip_make_real_var(name, net);
@@ -45,7 +46,8 @@ static void __compile_var_real(char*label, char*name,
       if (name) {
 	    assert(!array);
 	    vpip_attach_to_current_scope(obj);
-	    schedule_init_vector(vvp_net_ptr_t(net,0), fun->real_value());
+            if (!vpip_peek_current_scope()->is_automatic)
+                  schedule_init_vector(vvp_net_ptr_t(net,0), fun->real_value());
       }
       if (array) {
 	    assert(!name);
@@ -81,6 +83,7 @@ static void __compile_var(char*label, char*name,
       vvp_net_t*node = new vvp_net_t;
 
       node->fun = vsig;
+      vpip_add_item_to_current_scope(vsig);
       define_functor_symbol(label, node);
 
       vpiHandle obj = 0;
@@ -96,7 +99,8 @@ static void __compile_var(char*label, char*name,
       if (name) {
 	    assert(!array);
 	    if (obj) vpip_attach_to_current_scope(obj);
-	    schedule_init_vector(vvp_net_ptr_t(node,0), vsig->vec4_value());
+            if (!vpip_peek_current_scope()->is_automatic)
+	          schedule_init_vector(vvp_net_ptr_t(node,0), vsig->vec4_value());
       }
 	// If this is an array word, then it does not have a name, and
 	// it is attached to the addressed array.