This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Fix ACATS failures in GCC 3.4
- From: Eric Botcazou <ebotcazou at act-europe dot fr>
- To: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Thu, 26 Feb 2004 14:29:49 +0100
- Subject: Re: [PATCH] Fix ACATS failures in GCC 3.4
- References: <10402261311.AA26807@vlsi1.ultra.nyu.edu>
> No, I think Roger's fix is sufficient and the real problem. The point
> of the code you're patching in fold is to find the simplest representation
> of the arguments so their form can be studied. It's a serious error in
> fold to place those in the tree without converting them first. The code
> he's patching makes that error.
OK, I see the disagreement here. I thought it was always safe to strip casts
that are deemed NOPs for a particular node.
> But as long as you're not losing information by stripping any
> sign-changing NOPs from the operand, it's safe to do so. The only case
> where information is lost is the comparison case (since the type of the
> comparison node means something else), not MIN, MAX, or right shift.
Here I don't see it anymore. Hence the question: is it safe or not safe to
strip these NOPs for MAX_EXPR, since you're saying that only the type of the
MAX_EXPR is supposed to matter?