This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[PATCH] Fix PR 15777, Fold does not always fold sub trees


I found this bug when I was playing around with my tree combiner
which needs many improvements which I do not have time for right
now.

Basically ((x&1) == 0) == 0 is folded to
(int)((unsigned int)x ^ 1 ^ 1) & 1 when it should just fold it
to be (x & 1) != 0 or (x & 1) depending on the resulting type.

So this patch makes fold, fold the new x ^ 1 expression which fixes
this bug.

OK? Bootstrapped on powerpc-apple-darwin with no regression.

Thanks,
Andrew Pinski



ChangeLog:
	* fold-const.c (fold_single_bit_test): Fold the x^1 expression.


Index: fold-const.c =================================================================== RCS file: /cvs/gcc/gcc/gcc/fold-const.c,v retrieving revision 1.417 diff -u -p -r1.417 fold-const.c --- fold-const.c 5 Jul 2004 16:44:17 -0000 1.417 +++ fold-const.c 7 Jul 2004 04:58:11 -0000 @@ -5850,8 +5849,8 @@ fold_single_bit_test (enum tree_code cod inner, size_int (bitnum));

       if (code == EQ_EXPR)
-	inner = build2 (BIT_XOR_EXPR, intermediate_type,
-			inner, integer_one_node);
+	inner = fold (build2 (BIT_XOR_EXPR, intermediate_type,
+			      inner, integer_one_node));

       /* Put the AND last so it can combine with more things.  */
       inner = build2 (BIT_AND_EXPR, intermediate_type,


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]