fold bug with multiply and cast and conditional

[Andrew Pinski]

On Dec 8, 2004, at 11:03 PM, Roger Sayle wrote:

> Best of luck debugging your problem, wherever it may be :>  My initial
> guess would be that your combiner is not correctly casting the types of
> the operands of the multiplication to the appropriate/result type.
> Just a guess, but this would explain why you can't reproduce it with
> any of GCC's language front-ends, which correctly preserve tree type
> safety.

Sorry about misattributing this problem to your patch.  I should looked
into the problem a little further.

It was due to the wrong type but not by my tree combiner.  The
bug is in fold_binary_op_with_conditional_arg still.  We take
the type of the conditional to make true/false_value instead of
using the type of the outer operation so we get a miss-matched type.

It also explains why I cannot reproduce it with the front-ends (well I
only tried the C based ones) but the front-ends do not produce a
straight up conversion to int from a condition, but instead they
(really convert) creates a COND_EXPR which means we don't hit this
part of the code also.

This is the patch which I am testing which should fix the bug, this is a
regression in a sense from 2.95.3 which did not have
fold_binary_op_with_conditional_arg.

Index: fold-const.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/fold-const.c,v
retrieving revision 1.482
diff -u -p -r1.482 fold-const.c
--- fold-const.c	30 Nov 2004 20:33:34 -0000	1.482
+++ fold-const.c	9 Dec 2004 05:29:08 -0000
@@ -5392,10 +5392,9 @@ fold_binary_op_with_conditional_arg (enu
      }
    else
      {
-      tree testtype = TREE_TYPE (cond);
        test = cond;
-      true_value = constant_boolean_node (true, testtype);
-      false_value = constant_boolean_node (false, testtype);
+      true_value = constant_boolean_node (true, type);
+      false_value = constant_boolean_node (false, type);
      }

    if (lhs == 0)


Thanks,
Andrew Pinski