Re: Algoritmic Complexity Attacks and 2.4.20 the dcache code

[Aaron Lehmann <aaronl@vitelus.com>]

On Fri, May 30, 2003 at 11:45:29PM -0700, David S. Miller wrote:
>    From: William Lee Irwin III <wli@holomorphy.com>
>    Date: Fri, 30 May 2003 23:41:38 -0700
>    
>    If it's literally that trivial I'll put digging around the machine
>    descriptions on my TODO list.
> 
> Look at TARGET_RTX_COSTS, thats where all of this happens.

Reading the code that handles this stuff (expmed.c) always cracks me up.


  /* We might want to refine this now that we have division-by-constant
     optimization.  Since expand_mult_highpart tries so many variants, it is
     not straightforward to generalize this.  Maybe we should make an array
     of possible modes in init_expmed?  Save this for GCC 2.7.  */
 
        /* We could just as easily deal with negative constants here,
           but it does not seem worth the trouble for GCC 2.6.  */

        /* This is extremely similar to the code for the unsigned case
           above.  For 2.7 we should merge these variants, but for
           2.6.1 I don't want to touch the code for unsigned since that
           get used in C.  The signed case will only be used by other
           languages (Ada).  */

Sometimes I wish the gcc code was tame enough for me to work on.