[PATCH] Fix c54a13b

[PATCH] Fix c54a13b

[Andrew Pinski]

The problem here is that types can have non constant min and max
values which cause an ICE in stmt.c where we just use
tree_int_cst_compare instead of checking to make sure that we have
INTEGER_CST.

OK? Bootstrapped and tested on powerpc-darwin (Ada's support
library still not able to build fully here, I just tested
that the testcase now passes).

Thanks,
Andrew Pinski

ChangeLog:
	* stmt.c (add_case_node): Make sure that we have integer
	constant before calling tree_int_cst_compare.

Index: stmt.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/stmt.c,v
retrieving revision 1.398
diff -u -p -r1.398 stmt.c
--- stmt.c	11 Oct 2004 16:11:29 -0000	1.398
+++ stmt.c	12 Oct 2004 22:19:31 -0000
@@ -2139,8 +2139,10 @@ add_case_node (struct case_node *head, t
    if (!high || tree_int_cst_equal (low, high))
      {
        /* If the simple case value is unreachable, ignore it.  */
-      if (tree_int_cst_compare (low, min_value) < 0
-	  || tree_int_cst_compare (low, max_value) > 0)
+      if ((TREE_CODE (min_value) == INTEGER_CST
+            && tree_int_cst_compare (low, min_value) < 0)
+	  || (TREE_CODE (max_value) == INTEGER_CST
+	      && tree_int_cst_compare (low, max_value) > 0))
  	return head;
        low = fold_convert (type, low);
        high = low;
@@ -2148,19 +2150,23 @@ add_case_node (struct case_node *head, t
    else
      {
        /* If the entire case range is unreachable, ignore it.  */
-      if (tree_int_cst_compare (high, min_value) < 0
-	  || tree_int_cst_compare (low, max_value) > 0)
+      if ((TREE_CODE (min_value) == INTEGER_CST
+            && tree_int_cst_compare (high, min_value) < 0)
+	  || (TREE_CODE (max_value) == INTEGER_CST
+	      && tree_int_cst_compare (low, max_value) > 0))
  	return head;

        /* If the lower bound is less than the index type's minimum
  	 value, truncate the range bounds.  */
-      if (tree_int_cst_compare (low, min_value) < 0)
+      if (TREE_CODE (min_value) == INTEGER_CST
+            && tree_int_cst_compare (low, min_value) < 0)
  	low = min_value;
        low = fold_convert (type, low);

        /* If the upper bound is greater than the index type's maximum
  	 value, truncate the range bounds.  */
-      if (tree_int_cst_compare (high, max_value) > 0)
+      if (TREE_CODE (max_value) == INTEGER_CST
+	  && tree_int_cst_compare (high, max_value) > 0)
  	high = max_value;
        high = fold_convert (type, high);
      }

Re: [PATCH] Fix c54a13b

[Roger Sayle]

On Tue, 12 Oct 2004, Andrew Pinski wrote:
> 	* stmt.c (add_case_node): Make sure that we have integer
> 	constant before calling tree_int_cst_compare.

This is OK for mainline.  Sorry for the inconvenience to the Ada
folks.  This code was based upon c-common.c's check_case_bounds
which obviously doesn't have to worry about Ada's non-constant
type bounds.

Roger
--

Re: [PATCH] Fix c54a13b

[Laurent GUERBY]

I confirm andrew patch fixes c54a13b on x86 and x86_64 linux.

Laurent

On Wed, 2004-10-13 at 05:15, Roger Sayle wrote:
> On Tue, 12 Oct 2004, Andrew Pinski wrote:
> > 	* stmt.c (add_case_node): Make sure that we have integer
> > 	constant before calling tree_int_cst_compare.
> 
> This is OK for mainline.  Sorry for the inconvenience to the Ada
> folks.  This code was based upon c-common.c's check_case_bounds
> which obviously doesn't have to worry about Ada's non-constant
> type bounds.
> 
> Roger
> --
> 
> 

Re: [PATCH] Fix c54a13b

[Richard Kenner]

    The problem here is that types can have non constant min and max
    values which cause an ICE in stmt.c where we just use
    tree_int_cst_compare instead of checking to make sure that we have
    INTEGER_CST.

Interesting.  I don't see that test failing on my runs.

Re: [PATCH] Fix c54a13b

[Andrew Pinski]

On Oct 12, 2004, at 7:14 PM, Richard Kenner wrote:

>     The problem here is that types can have non constant min and max
>     values which cause an ICE in stmt.c where we just use
>     tree_int_cst_compare instead of checking to make sure that we have
>     INTEGER_CST.
>
> Interesting.  I don't see that test failing on my runs.

Laurent Guerby reported this to me on IRC and I decided to look into
as I was trying to figure out why the bootstrap is failing on
powerpc-darwin with Ada enabled.

Also the patch which caused this was committed yesterday:
2004-10-11  Roger Sayle  <roger@eyesopen.com>

         PR middle-end/17657
         * stmt.c (add_case_node): Add additional type argument.  Declare
         as static to match prototype.  Convert the upper and lower 
bounds
         to the specified index type.  Optimize away case ranges/values
         that are outside the index type's bounds.  Truncate case ranges
         that span the index type's bounds.
         (expand_case): Avoid unnessary computation and memory allocation
         when index type is error_mark_node.  Pass index_type as required
         by change to add_case_node API.  No need to convert case range
         bounds to index_type, this is now done by add_case_node.


Thanks,
Andrew Pinski

Re: [PATCH] Fix c54a13b

[Richard Kenner]

    Also the patch which caused this was committed yesterday:

That explains it.  My last ACATS run was 11 AM Sunday.