------- Comment #12 from law at redhat dot com 2006-06-08 16:38 -------
Subject: Re: bootstrap failure building libdecnumber,
ICE in compare_values, tree-vrp.c:432
On Thu, 2006-06-08 at 09:47 +0000, rguenth at gcc dot gnu dot org wrote:
>
> ------- Comment #11 from rguenth at gcc dot gnu dot org 2006-06-08 09:47
> -------
> This one:
>
> /* Apply the operation to each end of the range and see what we end
> up with. */
> if (code == NEGATE_EXPR
> && !TYPE_UNSIGNED (TREE_TYPE (expr)))
> {
> /* NEGATE_EXPR flips the range around. */
> min = (vr0.max == TYPE_MAX_VALUE (TREE_TYPE (expr)) && !flag_wrapv)
> ? TYPE_MIN_VALUE (TREE_TYPE (expr))
> : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max);
>
> max = (vr0.min == TYPE_MIN_VALUE (TREE_TYPE (expr)) && !flag_wrapv)
> ? TYPE_MAX_VALUE (TREE_TYPE (expr))
> : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.min);
>
> }
>
> is wrong for
>
> (gdb) print vr0
> $3 = {type = VR_ANTI_RANGE, min = 0xb7d2b3a8, max = 0xb7d2b3a8,
> equiv = 0x896c298}
> (gdb) call debug_generic_expr (vr0.min)
> -2147483648
> (gdb) call debug_generic_expr (vr0.max)
> -2147483648
>
> It needs to read:
>
> min = (vr0.max == TYPE_MIN_VALUE (TREE_TYPE (expr)) && !flag_wrapv)
> ? TYPE_MAX_VALUE (TREE_TYPE (expr))
> : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max);
>
> values other than TYPE_MIN_VALUE are not special with inversion.
I'm not terribly familiar with the code in question or this PR. But it
seems to me that negating a signed type's minimum value should give you
the same value with an overflow.
ie, for a 32bit signed type
- (-2147483648) => -2147483648
If I read your proposed change, you'd instead return the type's max
value, in this case 2147483647. Which seems rather wrong.
Diego is far more familiar with code than I am.
jeff
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26998
