calls.c (compute_argument_block_size): New argument preferred_stack_boundary.

* calls.c (compute_argument_block_size): New argument
	preferred_stack_boundary.
	(expand_call): update cfun->preferred_stack_boundary, update call of
	compute_argument_block_size
	(emit_library_call): Increate cfun->preferred_stack_boundary
	to PREFERRED_STACK_BOUNDARY
	(emit_library_call_value): Likewise.
	* explow.c (allocate_dynamic_stack_spave): Likewise.
	* function.c (prepare_function_start): Set
	cfun->preferred_stack_boundary
	* function.h (struct function): Add preferred_stack_boundary field.
	* integrate.c (expand_inline_function): Update
	cfun->preferred_stack_boundary and cfun->stack_alignment_needed.
	(copy_rtx_and_substitute): Align frame to stack_alignment_needed only.
	* i386.c (compute_frame_size): Use cfun->preferred_stack_boundary.

From-SVN: r31831

gcc/ChangeLog

@@ -1,3 +1,21 @@
+Mon Feb  7 18:36:41 MET 2000  Jan Hubicka  <jh@suse.cz>
+
+	* calls.c (compute_argument_block_size): New argument
+	preferred_stack_boundary.
+	(expand_call): update cfun->preferred_stack_boundary, update call of
+	compute_argument_block_size
+	(emit_library_call): Increate cfun->preferred_stack_boundary
+	to PREFERRED_STACK_BOUNDARY
+	(emit_library_call_value): Likewise.
+	* explow.c (allocate_dynamic_stack_spave): Likewise.
+	* function.c (prepare_function_start): Set
+	cfun->preferred_stack_boundary
+	* function.h (struct function): Add preferred_stack_boundary field.
+	* integrate.c (expand_inline_function): Update
+	cfun->preferred_stack_boundary and cfun->stack_alignment_needed.
+	(copy_rtx_and_substitute): Align frame to stack_alignment_needed only.
+	* i386.c (compute_frame_size): Use cfun->preferred_stack_boundary.
+
 2000-02-06  Zack Weinberg  <zack@wolery.cumb.org>
 
 	* cpplib.c (my_strerror, cpp_error, cpp_error_with_line,

gcc/calls.c

@@ -148,7 +148,8 @@ static void precompute_arguments 		PARAMS ((int, int, int,
 							 struct arg_data *,
 							 struct args_size *));
 static int compute_argument_block_size		PARAMS ((int, 
-							 struct args_size *));
+							 struct args_size *,
+							 int));
 static void initialize_argument_information	PARAMS ((int,
 							 struct arg_data *,
 							 struct args_size *,
@@ -1176,9 +1177,11 @@ initialize_argument_information (num_actuals, args, args_size, n_named_args,
    for arguments passed in registers.  */
 
 static int
-compute_argument_block_size (reg_parm_stack_space, args_size)
+compute_argument_block_size (reg_parm_stack_space, args_size,
+    			     preferred_stack_boundary)
      int reg_parm_stack_space;
      struct args_size *args_size;
+     int preferred_stack_boundary ATTRIBUTE_UNUSED;
 {
   int unadjusted_args_size = args_size->constant;
 
@@ -1192,8 +1195,9 @@ compute_argument_block_size (reg_parm_stack_space, args_size)
       args_size->constant = 0;
 
 #ifdef PREFERRED_STACK_BOUNDARY
-      if (PREFERRED_STACK_BOUNDARY != BITS_PER_UNIT)
+      preferred_stack_boundary /= BITS_PER_UNIT;
-	args_size->var = round_up (args_size->var, STACK_BYTES);
+      if (preferred_stack_boundary > 1)
+	args_size->var = round_up (args_size->var, preferred_stack_boundary);
 #endif
 
       if (reg_parm_stack_space > 0)
@@ -1214,10 +1218,12 @@ compute_argument_block_size (reg_parm_stack_space, args_size)
   else
     {
 #ifdef PREFERRED_STACK_BOUNDARY
+      preferred_stack_boundary /= BITS_PER_UNIT;
       args_size->constant = (((args_size->constant
 			       + pending_stack_adjust
-			       + STACK_BYTES - 1)
+			       + preferred_stack_boundary - 1)
-			      / STACK_BYTES * STACK_BYTES)
+			      / preferred_stack_boundary
+			      * preferred_stack_boundary)
 			     - pending_stack_adjust);
 #endif
 
@@ -1692,6 +1698,14 @@ expand_call (exp, target, ignore)
   rtx call_fusage = 0;
   register tree p;
   register int i;
+#ifdef PREFERRED_STACK_BOUNDARY
+  int preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
+#else
+  /* In this case preferred_stack_boundary variable is meaningless.
+     It is used only in order to keep ifdef noise down when calling
+     compute_argument_block_size.  */
+  int preferred_stack_boundary = 0;
+#endif
 
   /* The value of the function call can be put in a hard register.  But
      if -fcheck-memory-usage, code which invokes functions (and thus
@@ -1921,6 +1935,13 @@ expand_call (exp, target, ignore)
   if (fndecl && DECL_NAME (fndecl))
     name = IDENTIFIER_POINTER (DECL_NAME (fndecl));
 
+  /* Ensure current function's preferred stack boundary is at least
+     what we need.  We don't have to increase alignment for recursive
+     functions.  */
+  if (cfun->preferred_stack_boundary < preferred_stack_boundary
+      && fndecl != current_function_decl)
+    cfun->preferred_stack_boundary = preferred_stack_boundary;
+
   /* See if this is a call to a function that can return more than once
      or a call to longjmp or malloc.  */
   special_function_p (fndecl, &returns_twice, &is_longjmp, &fork_or_exec,
@@ -2062,7 +2083,8 @@ expand_call (exp, target, ignore)
      and constant sizes must be combined, the size may have to be rounded,
      and there may be a minimum required size.  */
   unadjusted_args_size
-    = compute_argument_block_size (reg_parm_stack_space, &args_size);
+    = compute_argument_block_size (reg_parm_stack_space, &args_size,
+				   preferred_stack_boundary);
 
   /* Now make final decision about preallocating stack space.  */
   must_preallocate = finalize_must_preallocate (must_preallocate,
@@ -2703,6 +2725,13 @@ emit_library_call VPARAMS((rtx orgfun, int no_queue, enum machine_mode outmode,
 
   push_temp_slots ();
 
+#ifdef PREFERRED_STACK_BOUNDARY
+  /* Ensure current function's preferred stack boundary is at least
+     what we need.  */
+  if (cfun->preferred_stack_boundary < PREFERRED_STACK_BOUNDARY)
+    cfun->preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
+#endif
+
   for (count = 0; count < nargs; count++)
     {
       rtx val = va_arg (p, rtx);
@@ -3192,6 +3221,13 @@ emit_library_call_value VPARAMS((rtx orgfun, rtx value, int no_queue,
   is_const = no_queue;
   fun = orgfun;
 
+#ifdef PREFERRED_STACK_BOUNDARY
+  /* Ensure current function's preferred stack boundary is at least
+     what we need.  */
+  if (cfun->preferred_stack_boundary < PREFERRED_STACK_BOUNDARY)
+    cfun->preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
+#endif
+
   /* If this kind of value comes back in memory,
      decide where in memory it should come back.  */
   if (aggregate_value_p (type_for_mode (outmode, 0)))

gcc/config/i386/i386.c

@@ -1776,7 +1776,7 @@ ix86_compute_frame_size (size, nregs_on_stack, rpadding1, rpadding2)
 #ifdef PREFERRED_STACK_BOUNDARY
   {
     int offset;
-    int preferred_alignment = PREFERRED_STACK_BOUNDARY / BITS_PER_UNIT;
+    int preferred_alignment = cfun->preferred_stack_boundary / BITS_PER_UNIT;
 
     offset = frame_pointer_needed ? 8 : 4;
 

gcc/explow.c

@@ -1178,6 +1178,13 @@ allocate_dynamic_stack_space (size, target, known_align)
   if (GET_MODE (size) != VOIDmode && GET_MODE (size) != Pmode)
     size = convert_to_mode (Pmode, size, 1);
 
+  /* We can't attempt to minimize alignment necessary, because we don't
+     know the final value of preferred_stack_boundary yet while executing
+     this code.  */
+#ifdef PREFERRED_STACK_BOUNDARY
+  cfun->preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
+#endif
+
   /* We will need to ensure that the address we return is aligned to
      BIGGEST_ALIGNMENT.  If STACK_DYNAMIC_OFFSET is defined, we don't
      always know its final value at this point in the compilation (it 

gcc/function.c

@@ -5686,6 +5686,9 @@ prepare_function_start ()
   cfun->original_arg_vector = 0;  
 
   cfun->stack_alignment_needed = 0;
+#ifdef STACK_BOUNDARY
+  cfun->preferred_stack_boundary = STACK_BOUNDARY;
+#endif
 
   /* Set if a call to setjmp is seen.  */
   current_function_calls_setjmp = 0;

gcc/function.h

@@ -456,6 +456,8 @@ struct function
   struct machine_function *machine;
   /* The largest alignment of slot allocated on the stack.  */
   int stack_alignment_needed;
+  /* Preferred alignment of the end of stack frame.  */
+  int preferred_stack_boundary;
 
   /* Language-specific code can use this to store whatever it likes.  */
   struct language_function *language;

gcc/integrate.c

@@ -602,6 +602,12 @@ expand_inline_function (fndecl, parms, target, ignore, type,
 
   nargs = list_length (DECL_ARGUMENTS (fndecl));
 
+  if (cfun->preferred_stack_boundary < inl_f->preferred_stack_boundary)
+    cfun->preferred_stack_boundary = inl_f->preferred_stack_boundary;
+
+  if (cfun->stack_alignment_needed < inl_f->stack_alignment_needed)
+    cfun->stack_alignment_needed = inl_f->stack_alignment_needed;
+
   /* Check that the parms type match and that sufficient arguments were
      passed.  Since the appropriate conversions or default promotions have
      already been applied, the machine modes should match exactly.  */
@@ -1619,13 +1625,17 @@ copy_rtx_and_substitute (orig, map, for_lhs)
 	    {
 	      rtx loc, seq;
 	      int size = get_func_frame_size (DECL_SAVED_INSNS (map->fndecl));
+	      int alignment
+		= (DECL_SAVED_INSNS (map->fndecl)->stack_alignment_needed
+		   / BITS_PER_UNIT);
 
 #ifdef FRAME_GROWS_DOWNWARD
 	      /* In this case, virtual_stack_vars_rtx points to one byte
 		 higher than the top of the frame area.  So make sure we
 		 allocate a big enough chunk to keep the frame pointer
 		 aligned like a real one.  */
-	      size = CEIL_ROUND (size, BIGGEST_ALIGNMENT / BITS_PER_UNIT);
+	      if (alignment)
+		size = CEIL_ROUND (size, alignment);
 #endif
 	      start_sequence ();
 	      loc = assign_stack_temp (BLKmode, size, 1);