PR middle-end/25121: [4.1/4.2 Regression] libgcj misscompilation? [Was Re: RFC: peephole vs RTX_FRAME_RELATED_P]

[Richard Henderson]

On Tue, Dec 20, 2005 at 11:40:20AM +0000, Andrew Haley wrote:
> +	      /* Don't replace sequences with FRAME_RELATED_P insns.  */
> +	      if (try != NULL && match_len < frame_related_distance)

I think I'd rather that you manipulate peep2_current_count
than adding a new counter.


r~