More i386 comparison fixes

[Richard Henderson]

On Tue, Oct 17, 2000 at 11:23:17AM +0200, Jan Hubicka wrote:
> Not exactly - it reverses only carry, not overflow.

Right.

> Also combine can't canonize always, even with my patch, since it must be
> able to swap the condition code, so it must see the "combined" CC setter
> and user patterns and often it is not able to split it back to matching
> instructions, so this scheme is unreliable and third, this reversed order
> is very common, since it comes from canonizing (compare (plus a b)
> (const_int 0)) type of expressions.

Ok.

The patch is approved, then.



r~