Re: [patch] VRP miscompiles crafty

[Geoff Keating <geoffk at geoffk dot org>]

On 13/09/2005, at 6:16 AM, Michael Matz wrote:

> Hi,
>
> On Mon, 11 Sep 2005, Geoffrey Keating wrote:
>
>
> > Although it's pretty clear that 0 - (-Inf) should be +Inf, I'm a bit
> > confused by how the code got to this point.  The expression '0 - (- 
> > Inf)'
> > does not overflow in itself;
>
> It does.  -INF == MIN_INT == -2147483648 (or rather TYPE_MIN_VALUE,  
> but
> you get the idea).  And the negative of that is not representable  
> in the
> same type, hence an overflow occurs when producing it.  +INF is
> TYPE_MAX_VALUE which is not -TYPE_MIN_VALUE.  Remember that we use  
> normal
> integers to represent -+INF, not any exceptional value, and normal  
> integer
> arithmetic (int_const_binop) to calculate with them.

OK.  In that case, I was completely confused (the references to 'INF'  
made me think this was floating-point), I will look at the patch again.

Attachment: smime.p7s
Description: S/MIME cryptographic signature