re PR rtl-optimization/33721 ([meta-bug] Gcc can't properly align stack variable)

	PR rtl-opt/33721
	* explow.c (allocate_dynamic_stack_space): Add REQUIRED_ALIGN parm,
	remove TARGET parm, convert KNOWN_ALIGN parm to SIZE_ALIGN.  Honor
	required_align, tidy the code a bit.  Emit split_stack code in the
	right place.  Mark the return value with the alignment properly.
	* expr.h (allocate_dynamic_stack_space): Update decl.
	* builtins.c (expand_builtin_apply): Update call to
	allocate_dynamic_stack_space.
	(expand_builtin_alloca): Likewise.  Remove TARGET parameter.
	* calls.c (initialize_argument_information): Update call to
	allocate_dynamic_stack_space.
	(expand_call): Likewise.

	* cfgexpand.c (get_decl_align_unit): Don't limit alignment.
	Don't update_stack_alignment here.
	(alloc_stack_frame_space): Make ALIGN unsigned.
	(stack_var_cmp): Sort by alignment too.
	(partition_stack_vars): Don't merge large and small alignment vars.
	(expand_one_stack_var_at): Add BASE and BASE_ALIGN parameters.
	Take care when BASE is not virtual_stack_vars_rtx.
	(expand_stack_vars): Allocate dynamic stack space for large
	alignment variables.
	(expand_one_stack_var): Update all to expand_one_stack_var_at.
	(defer_stack_allocation): True for large alignment vars.
	(update_stack_alignment): Merge into ...
	(expand_one_var): ... here.
	(gimple_expand_cfg): Place code from expand_stack_vars.

From-SVN: r165240

gcc/ChangeLog

+2010-10-09  Richard Henderson  <rth@redhat.com>
+
+	PR rtl-opt/33721
+	* explow.c (allocate_dynamic_stack_space): Add REQUIRED_ALIGN parm,
+	remove TARGET parm, convert KNOWN_ALIGN parm to SIZE_ALIGN.  Honor
+	required_align, tidy the code a bit.  Emit split_stack code in the
+	right place.  Mark the return value with the alignment properly.
+	* expr.h (allocate_dynamic_stack_space): Update decl.
+	* builtins.c (expand_builtin_apply): Update call to
+	allocate_dynamic_stack_space.
+	(expand_builtin_alloca): Likewise.  Remove TARGET parameter.
+	* calls.c (initialize_argument_information): Update call to
+	allocate_dynamic_stack_space.
+	(expand_call): Likewise.
+
+	* cfgexpand.c (get_decl_align_unit): Don't limit alignment.
+	Don't update_stack_alignment here.
+	(alloc_stack_frame_space): Make ALIGN unsigned.
+	(stack_var_cmp): Sort by alignment too.
+	(partition_stack_vars): Don't merge large and small alignment vars.
+	(expand_one_stack_var_at): Add BASE and BASE_ALIGN parameters.
+	Take care when BASE is not virtual_stack_vars_rtx.
+	(expand_stack_vars): Allocate dynamic stack space for large
+	alignment variables.
+	(expand_one_stack_var): Update all to expand_one_stack_var_at.
+	(defer_stack_allocation): True for large alignment vars.
+	(update_stack_alignment): Merge into ...
+	(expand_one_var): ... here.
+	(gimple_expand_cfg): Place code from expand_stack_vars.
+
 2010-10-09  Nathan Froyd  <froydnj@codesourcery.com>
 
 	* config/pdp11/pdp11.h (FUNCTION_ARG, FUNCTION_ARG_ADVANCE): Delete.

gcc/builtins.c

@@ -132,7 +132,7 @@ static rtx expand_builtin_memset (tree, rtx, enum machine_mode);
 static rtx expand_builtin_memset_args (tree, tree, tree, rtx, enum machine_mode, tree);
 static rtx expand_builtin_bzero (tree);
 static rtx expand_builtin_strlen (tree, rtx, enum machine_mode);
-static rtx expand_builtin_alloca (tree, rtx, bool);
+static rtx expand_builtin_alloca (tree, bool);
 static rtx expand_builtin_unop (enum machine_mode, tree, rtx, rtx, optab);
 static rtx expand_builtin_frame_address (tree, tree);
 static tree stabilize_va_list_loc (location_t, tree, int);
@@ -1572,7 +1572,7 @@ expand_builtin_apply (rtx function, rtx arguments, rtx argsize)
      arguments to the outgoing arguments address.  We can pass TRUE
      as the 4th argument because we just saved the stack pointer
      and will restore it right after the call.  */
-  allocate_dynamic_stack_space (argsize, 0, BITS_PER_UNIT, TRUE);
+  allocate_dynamic_stack_space (argsize, 0, BIGGEST_ALIGNMENT, true);
 
   /* Set DRAP flag to true, even though allocate_dynamic_stack_space
      may have already set current_function_calls_alloca to true.
@@ -4931,12 +4931,11 @@ expand_builtin_frame_address (tree fndecl, tree exp)
 }
 
 /* Expand EXP, a call to the alloca builtin.  Return NULL_RTX if we
-   failed and the caller should emit a normal call, otherwise try to
-   get the result in TARGET, if convenient.  CANNOT_ACCUMULATE is the
-   same as for allocate_dynamic_stack_space.  */
+   failed and the caller should emit a normal call.  CANNOT_ACCUMULATE
+   is the same as for allocate_dynamic_stack_space.  */
 
 static rtx
-expand_builtin_alloca (tree exp, rtx target, bool cannot_accumulate)
+expand_builtin_alloca (tree exp, bool cannot_accumulate)
 {
   rtx op0;
   rtx result;
@@ -4952,7 +4951,7 @@ expand_builtin_alloca (tree exp, rtx target, bool cannot_accumulate)
   op0 = expand_normal (CALL_EXPR_ARG (exp, 0));
 
   /* Allocate the desired space.  */
-  result = allocate_dynamic_stack_space (op0, target, BITS_PER_UNIT,
+  result = allocate_dynamic_stack_space (op0, 0, BIGGEST_ALIGNMENT,
 					 cannot_accumulate);
   result = convert_memory_address (ptr_mode, result);
 
@@ -5997,7 +5996,7 @@ expand_builtin (tree exp, rtx target, rtx subtarget, enum machine_mode mode,
     case BUILT_IN_ALLOCA:
       /* If the allocation stems from the declaration of a variable-sized
 	 object, it cannot accumulate.  */
-      target = expand_builtin_alloca (exp, target, ALLOCA_FOR_VAR_P (exp));
+      target = expand_builtin_alloca (exp, ALLOCA_FOR_VAR_P (exp));
       if (target)
 	return target;
       break;

gcc/calls.c

@@ -1100,10 +1100,11 @@ initialize_argument_information (int num_actuals ATTRIBUTE_UNUSED,
 		  /* We can pass TRUE as the 4th argument because we just
 		     saved the stack pointer and will restore it right after
 		     the call.  */
-		  copy = gen_rtx_MEM (BLKmode,
-				      allocate_dynamic_stack_space
-				      (size_rtx, NULL_RTX,
-				       TYPE_ALIGN (type), TRUE));
+		  copy = allocate_dynamic_stack_space (size_rtx,
+						       TYPE_ALIGN (type),
+						       TYPE_ALIGN (type),
+						       true);
+		  copy = gen_rtx_MEM (BLKmode, copy);
 		  set_mem_attributes (copy, type, 1);
 		}
 	      else
@@ -2664,8 +2665,8 @@ expand_call (tree exp, rtx target, int ignore)
 	      /* We can pass TRUE as the 4th argument because we just
 		 saved the stack pointer and will restore it right after
 		 the call.  */
-	      allocate_dynamic_stack_space (push_size, NULL_RTX,
-					    BITS_PER_UNIT, TRUE);
+	      allocate_dynamic_stack_space (push_size, 0,
+					    BIGGEST_ALIGNMENT, true);
 	    }
 
 	  /* If argument evaluation might modify the stack pointer,

gcc/explow.c

@@ -1123,15 +1123,18 @@ update_nonlocal_goto_save_area (void)
 }
 
 /* Return an rtx representing the address of an area of memory dynamically
-   pushed on the stack.  This region of memory is always aligned to
-   a multiple of BIGGEST_ALIGNMENT.
+   pushed on the stack.
 
    Any required stack pointer alignment is preserved.
 
    SIZE is an rtx representing the size of the area.
-   TARGET is a place in which the address can be placed.
 
-   KNOWN_ALIGN is the alignment (in bits) that we know SIZE has.
+   SIZE_ALIGN is the alignment (in bits) that we know SIZE has.  This
+   parameter may be zero.  If so, a proper value will be extracted 
+   from SIZE if it is constant, otherwise BITS_PER_UNIT will be assumed.
+
+   REQUIRED_ALIGN is the alignment (in bits) required for the region
+   of memory.
 
    If CANNOT_ACCUMULATE is set to TRUE, the caller guarantees that the
    stack space allocated by the generated code cannot be added with itself
@@ -1141,12 +1144,12 @@ update_nonlocal_goto_save_area (void)
    loops to it executes the associated deallocation code.  */
 
 rtx
-allocate_dynamic_stack_space (rtx size, rtx target, int known_align,
-			      bool cannot_accumulate)
+allocate_dynamic_stack_space (rtx size, unsigned size_align,
+			      unsigned required_align, bool cannot_accumulate)
 {
   HOST_WIDE_INT stack_usage_size = -1;
-  bool known_align_valid = true;
-  rtx final_label, final_target;
+  rtx final_label, final_target, target;
+  bool must_align;
 
   /* If we're asking for zero bytes, it doesn't matter what we point
      to since we can't dereference it.  But return a reasonable
@@ -1192,6 +1195,23 @@ allocate_dynamic_stack_space (rtx size, rtx target, int known_align,
   if (GET_MODE (size) != VOIDmode && GET_MODE (size) != Pmode)
     size = convert_to_mode (Pmode, size, 1);
 
+  /* Adjust SIZE_ALIGN, if needed.  */
+  if (CONST_INT_P (size))
+    {
+      unsigned HOST_WIDE_INT lsb;
+
+      lsb = INTVAL (size);
+      lsb &= -lsb;
+
+      /* Watch out for overflow truncating to "unsigned".  */
+      if (lsb > UINT_MAX / BITS_PER_UNIT)
+	size_align = 1u << (HOST_BITS_PER_INT - 1);
+      else
+	size_align = (unsigned)lsb * BITS_PER_UNIT;
+    }
+  else if (size_align < BITS_PER_UNIT)
+    size_align = BITS_PER_UNIT;
+
   /* 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.  */
@@ -1199,35 +1219,43 @@ allocate_dynamic_stack_space (rtx size, rtx target, int known_align,
     crtl->preferred_stack_boundary = PREFERRED_STACK_BOUNDARY;
 
   /* We will need to ensure that the address we return is aligned to
-     BIGGEST_ALIGNMENT.  If STACK_DYNAMIC_OFFSET is defined, we don't
+     REQUIRED_ALIGN.  If STACK_DYNAMIC_OFFSET is defined, we don't
      always know its final value at this point in the compilation (it
      might depend on the size of the outgoing parameter lists, for
      example), so we must align the value to be returned in that case.
      (Note that STACK_DYNAMIC_OFFSET will have a default nonzero value if
      STACK_POINTER_OFFSET or ACCUMULATE_OUTGOING_ARGS are defined).
      We must also do an alignment operation on the returned value if
-     the stack pointer alignment is less strict that BIGGEST_ALIGNMENT.
+     the stack pointer alignment is less strict than REQUIRED_ALIGN.
 
      If we have to align, we must leave space in SIZE for the hole
      that might result from the alignment operation.  */
 
+  must_align = (crtl->preferred_stack_boundary < required_align);
 #if defined (STACK_DYNAMIC_OFFSET) || defined (STACK_POINTER_OFFSET)
-#define MUST_ALIGN 1
-#else
-#define MUST_ALIGN (crtl->preferred_stack_boundary < BIGGEST_ALIGNMENT)
+  must_align = true;
 #endif
 
-  if (MUST_ALIGN)
+  if (must_align)
     {
-      size
-        = force_operand (plus_constant (size,
-					BIGGEST_ALIGNMENT / BITS_PER_UNIT - 1),
-			 NULL_RTX);
+      unsigned extra, extra_align;
+
+      if (required_align > PREFERRED_STACK_BOUNDARY)
+	extra_align = PREFERRED_STACK_BOUNDARY;
+      else if (required_align > STACK_BOUNDARY)
+	extra_align = STACK_BOUNDARY;
+      else
+	extra_align = BITS_PER_UNIT;
+      extra = (required_align - extra_align) / BITS_PER_UNIT;
+
+      size = plus_constant (size, extra);
+      size = force_operand (size, NULL_RTX);
 
       if (flag_stack_usage)
-	stack_usage_size += BIGGEST_ALIGNMENT / BITS_PER_UNIT - 1;
+	stack_usage_size += extra;
 
-      known_align_valid = false;
+      if (extra && size_align > extra_align)
+	size_align = extra_align;
     }
 
 #ifdef SETJMP_VIA_SAVE_AREA
@@ -1257,7 +1285,8 @@ allocate_dynamic_stack_space (rtx size, rtx target, int known_align,
       if (flag_stack_usage)
 	current_function_dynamic_alloc_count++;
 
-      known_align_valid = false;
+      /* ??? Can we infer a minimum of STACK_BOUNDARY here?  */
+      size_align = BITS_PER_UNIT;
     }
 #endif /* SETJMP_VIA_SAVE_AREA */
 
@@ -1274,7 +1303,7 @@ allocate_dynamic_stack_space (rtx size, rtx target, int known_align,
      insns.  Since this is an extremely rare event, we have no reliable
      way of knowing which systems have this problem.  So we avoid even
      momentarily mis-aligning the stack.  */
-  if (!known_align_valid || known_align % MAX_SUPPORTED_STACK_ALIGNMENT != 0)
+  if (size_align % MAX_SUPPORTED_STACK_ALIGNMENT != 0)
     {
       size = round_push (size);
 
@@ -1285,13 +1314,7 @@ allocate_dynamic_stack_space (rtx size, rtx target, int known_align,
 	}
     }
 
-  /* Don't use a TARGET that isn't a pseudo or is the wrong mode.  */
-  if (target == 0 || !REG_P (target)
-      || REGNO (target) < FIRST_PSEUDO_REGISTER
-      || GET_MODE (target) != Pmode)
-    target = gen_reg_rtx (Pmode);
-
-  mark_reg_pointer (target, known_align);
+  target = gen_reg_rtx (Pmode);
 
   /* The size is supposed to be fully adjusted at this point so record it
      if stack usage info is requested.  */
@@ -1341,7 +1364,6 @@ allocate_dynamic_stack_space (rtx size, rtx target, int known_align,
 	return space;
 
       final_target = gen_reg_rtx (Pmode);
-      mark_reg_pointer (final_target, known_align);
 
       emit_move_insn (final_target, space);
 
@@ -1440,35 +1462,38 @@ allocate_dynamic_stack_space (rtx size, rtx target, int known_align,
 #endif
     }
 
-  if (MUST_ALIGN)
+  /* Finish up the split stack handling.  */
+  if (final_label != NULL_RTX)
+    {
+      gcc_assert (flag_split_stack);
+      emit_move_insn (final_target, target);
+      emit_label (final_label);
+      target = final_target;
+    }
+
+  if (must_align)
     {
       /* CEIL_DIV_EXPR needs to worry about the addition overflowing,
 	 but we know it can't.  So add ourselves and then do
 	 TRUNC_DIV_EXPR.  */
       target = expand_binop (Pmode, add_optab, target,
-			     GEN_INT (BIGGEST_ALIGNMENT / BITS_PER_UNIT - 1),
+			     GEN_INT (required_align / BITS_PER_UNIT - 1),
 			     NULL_RTX, 1, OPTAB_LIB_WIDEN);
       target = expand_divmod (0, TRUNC_DIV_EXPR, Pmode, target,
-			      GEN_INT (BIGGEST_ALIGNMENT / BITS_PER_UNIT),
+			      GEN_INT (required_align / BITS_PER_UNIT),
 			      NULL_RTX, 1);
       target = expand_mult (Pmode, target,
-			    GEN_INT (BIGGEST_ALIGNMENT / BITS_PER_UNIT),
+			    GEN_INT (required_align / BITS_PER_UNIT),
 			    NULL_RTX, 1);
     }
 
+  /* Now that we've committed to a return value, mark its alignment.  */
+  mark_reg_pointer (target, required_align);
+
   /* Record the new stack level for nonlocal gotos.  */
   if (cfun->nonlocal_goto_save_area != 0)
     update_nonlocal_goto_save_area ();
 
-  /* Finish up the split stack handling.  */
-  if (final_label != NULL_RTX)
-    {
-      gcc_assert (flag_split_stack);
-      emit_move_insn (final_target, target);
-      emit_label (final_label);
-      target = final_target;
-    }
-
   return target;
 }
 

gcc/expr.h

@@ -642,7 +642,7 @@ extern void emit_stack_restore (enum save_level, rtx, rtx);
 extern void update_nonlocal_goto_save_area (void);
 
 /* Allocate some space on the stack dynamically and return its address.  */
-extern rtx allocate_dynamic_stack_space (rtx, rtx, int, bool);
+extern rtx allocate_dynamic_stack_space (rtx, unsigned, unsigned, bool);
 
 /* Emit one stack probe at ADDRESS, an address within the stack.  */
 extern void emit_stack_probe (rtx);

gcc/testsuite/ChangeLog

+2010-10-09  Richard Henderson  <rth@redhat.com>
+
+	* lib/target-supports.exp
+	(check_effective_target_automatic_stack_alignment): Always true.
+	
 2010-10-09  Richard Guenther  <rguenther@suse.de>
 
 	PR lto/45956

gcc/testsuite/lib/target-supports.exp

@@ -3551,12 +3551,8 @@ proc check_effective_target_4byte_wchar_t { } {
 # Return 1 if the target supports automatic stack alignment.
 
 proc check_effective_target_automatic_stack_alignment  { } {
-    if { [istarget i?86*-*-*]
-	 || [istarget x86_64-*-*] } then {
-	return 1
-    } else {
-	return 0
-    }
+    # Not "stack alignment" per se, but proper stack alignment of decls.
+    return 1;
 }
 
 # Return 1 if avx instructions can be compiled.