Re: [PATCH] Improvements to tree_expr_nonnegative_p

["Joseph S. Myers" <jsm28 at cam dot ac dot uk>]

On Sat, 7 Jun 2003, Roger Sayle wrote:

> addition and multiplication.  It looks like tree_expr_nonnegative_p
> was originally designed to primarily handle integer expressions, where
> the perils of overflow means that x*y may be negative even though both
> x and y are non-negative.  Floating point arithmetic on the other hand
> doesn't suffer from this problem.

Unless -fwrapv, neither does integer arithmetic.  That way you can fix bug
168, the oldest open non-C++ bug.  I pointed this out in a message I sent
to that bug (below), but Bugzilla appears to have silently lost that
message (it didn't get to gcc-bugs, I didn't get a bounce), although
messages I sent to 2493 the previous day and 704 on the 28th are recorded
fine.

-- 
Joseph S. Myers
jsm28@cam.ac.uk

Date: Mon, 26 May 2003 19:38:01 +0100 (BST)
From: Joseph S. Myers <jsm28@cam.ac.uk>
To: "pinskia@physics.uc.edu" <gcc-bugzilla@gcc.gnu.org>
Subject: Re: [Bug middle-end/168] Spurious signed/unsigned comparison
    warning

On Mon, 26 May 2003, pinskia@physics.uc.edu wrote:

> still happens on the mainline (20030526).

I think that this bug - the oldest open non-C++ bug - should actually be
easily fixable, and does not need value range propagation as the comments
suggest.

In order to avoid warning for the code given:

 int main () {
   unsigned char      uc;
   unsigned short int usi;
   unsigned int       ui;
   if (uc + usi >= ui);
 }

you just need to know that signed arithmetic overflow is undefined, so
either the LHS is nonnegative or undefined behavior has occurred.  Thus
tree_expr_nonnegative_p needs to know that unsigned values are
nonnegative, including when converted to wider signed values, and that the
sum or product of two nonnegative values is nonnegative.

-- 
Joseph S. Myers
jsm28@cam.ac.uk