diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 933b3b1761e..dbcdf3e26cc 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,13 @@
+2003-11-20  Fariborz Jahanian  <fjahanian@apple.com>
+	    David Edelsohn  <edelsohn@gnu.org>
+
+	* calls.c (expand_call): Allocate new temp in pass1.
+	(store_one_arg): If PARALLEL, calculate excess using mode size of
+	rtvec elt. 
+	* expr.c (emit_push_insn): If PARALLEL, calculate offset using
+	mode size of rtvec elt.
+	* function.c (assign_parms): Use parm in register, if available.
+
 2003-11-20  Kazu Hirata  <kazu@cs.umass.edu>
 
 	* config/h8300/lib1funcs.asm (MOVP): Remove.
diff --git a/gcc/calls.c b/gcc/calls.c
index a3b86ade979..79d9b9c4452 100644
--- a/gcc/calls.c
+++ b/gcc/calls.c
@@ -2152,6 +2152,7 @@ expand_call (tree exp, rtx target, int ignore)
 #endif
 
   int initial_highest_arg_in_use = highest_outgoing_arg_in_use;
+  rtx temp_target = 0;
   char *initial_stack_usage_map = stack_usage_map;
 
   int old_stack_allocated;
@@ -3266,7 +3267,11 @@ expand_call (tree exp, rtx target, int ignore)
 	 The Irix 6 ABI has examples of this.  */
       else if (GET_CODE (valreg) == PARALLEL)
 	{
-	  if (target == 0)
+	  /* Second condition is added because "target" is freed at the
+	     the end of "pass0" for -O2 when call is made to
+	     expand_end_target_temps ().  Its "in_use" flag has been set
+	     to false, so allocate a new temp.  */
+	  if (target == 0 || (pass == 1 && target == temp_target))
 	    {
 	      /* This will only be assigned once, so it can be readonly.  */
 	      tree nt = build_qualified_type (TREE_TYPE (exp),
@@ -3274,6 +3279,7 @@ expand_call (tree exp, rtx target, int ignore)
 					       | TYPE_QUAL_CONST));
 
 	      target = assign_temp (nt, 0, 1, 1);
+	      temp_target = target;
 	      preserve_temp_slots (target);
 	    }
 
@@ -4615,9 +4621,18 @@ store_one_arg (struct arg_data *arg, rtx argblock, int flags,
 	{
 	  /* PUSH_ROUNDING has no effect on us, because
 	     emit_push_insn for BLKmode is careful to avoid it.  */
-	  excess = (arg->locate.size.constant
-		    - int_size_in_bytes (TREE_TYPE (pval))
-		    + partial * UNITS_PER_WORD);
+	  if (reg && GET_CODE (reg) == PARALLEL)
+	  {
+	    /* Use the size of the elt to compute excess.  */
+	    rtx elt = XEXP (XVECEXP (reg, 0, 0), 0);
+	    excess = (arg->locate.size.constant
+		      - int_size_in_bytes (TREE_TYPE (pval))
+		      + partial * GET_MODE_SIZE (GET_MODE (elt)));
+	  } 
+	  else
+	    excess = (arg->locate.size.constant
+		      - int_size_in_bytes (TREE_TYPE (pval))
+		      + partial * UNITS_PER_WORD);
 	  size_rtx = expand_expr (size_in_bytes (TREE_TYPE (pval)),
 				  NULL_RTX, TYPE_MODE (sizetype), 0);
 	}
diff --git a/gcc/expr.c b/gcc/expr.c
index 320284c99ff..ac0249f3881 100644
--- a/gcc/expr.c
+++ b/gcc/expr.c
@@ -3474,9 +3474,19 @@ emit_push_insn (rtx x, enum machine_mode mode, tree type, rtx size,
 
       rtx temp;
       int used = partial * UNITS_PER_WORD;
-      int offset = used % (PARM_BOUNDARY / BITS_PER_UNIT);
+      int offset;
       int skip;
 
+      if (reg && GET_CODE (reg) == PARALLEL)
+	{
+	  /* Use the size of the elt to compute offset.  */
+	  rtx elt = XEXP (XVECEXP (reg, 0, 0), 0);
+	  used = partial * GET_MODE_SIZE (GET_MODE (elt));
+	  offset = used % (PARM_BOUNDARY / BITS_PER_UNIT);
+	}
+      else
+	offset = used % (PARM_BOUNDARY / BITS_PER_UNIT);
+
       if (size == 0)
 	abort ();
 
diff --git a/gcc/function.c b/gcc/function.c
index 04d34243dff..7bd31906a54 100644
--- a/gcc/function.c
+++ b/gcc/function.c
@@ -4704,6 +4704,15 @@ assign_parms (tree fndecl)
 
 	 Set DECL_RTL to that place.  */
 
+      if (GET_CODE (entry_parm) == PARALLEL && nominal_mode != BLKmode)
+	{
+	  /* Objects the size of a register can be combined in registers */
+	  rtx parmreg = gen_reg_rtx (nominal_mode);
+	  emit_group_store (parmreg, entry_parm, TREE_TYPE (parm),
+			    int_size_in_bytes (TREE_TYPE (parm)));
+	  SET_DECL_RTL (parm, parmreg);
+	}
+
       if (nominal_mode == BLKmode
 #ifdef BLOCK_REG_PADDING
 	  || (locate.where_pad == (BYTES_BIG_ENDIAN ? upward : downward)
@@ -4727,7 +4736,8 @@ assign_parms (tree fndecl)
 		 assign_stack_local if space was not allocated in the argument
 		 list.  If it was, this will not work if PARM_BOUNDARY is not
 		 a multiple of BITS_PER_WORD.  It isn't clear how to fix this
-		 if it becomes a problem.  */
+		 if it becomes a problem.  Exception is when BLKmode arrives
+		 with arguments not conforming to word_mode.  */
 
 	      if (stack_parm == 0)
 		{
@@ -4735,7 +4745,9 @@ assign_parms (tree fndecl)
 		  PUT_MODE (stack_parm, GET_MODE (entry_parm));
 		  set_mem_attributes (stack_parm, parm, 1);
 		}
-
+	      else if (GET_CODE (entry_parm) == PARALLEL 
+		       && GET_MODE(entry_parm) == BLKmode)
+		;
 	      else if (PARM_BOUNDARY % BITS_PER_WORD != 0)
 		abort ();
 
@@ -4798,7 +4810,10 @@ assign_parms (tree fndecl)
 		move_block_from_reg (REGNO (entry_parm), mem,
 				     size_stored / UNITS_PER_WORD);
 	    }
-	  SET_DECL_RTL (parm, stack_parm);
+	  /* If parm is already bound to register pair, don't change 
+	     this binding. */
+	  if (! DECL_RTL_SET_P (parm))
+	    SET_DECL_RTL (parm, stack_parm);
 	}
       else if (! ((! optimize
 		   && ! DECL_REGISTER (parm))