[4.0 PATCH] Fix PR middle-end/20177

[Roger Sayle]

On Sun, 20 Mar 2005, Mostafa Hagog wrote:
> 2005-03-20 Mostafa Hagog <mustafa@il.ibm.com>
>
>       PR middle-end/20177
>       * ddg.c (create_ddg_dependence): Ignore reg-anti dependency.
>       * modulo-sched.c (const_iteration_count): Return on NULL
>       pre-header.
>       (print_node_sched_params): Return on NULL dump_file.
>       (generate_reg_moves): Handle reg-anti dependencies and disregard
>       closing branch when generating register moves.
>       (sms_schedule): Mark the SMSed block dirty.
>       * passes.c (rest_of_handle_sms): Call update_life_info for all
>       basic-blocks.

This is OK for the 4.0 branch after it's been on mainline for two or
three days without problems.  I notice many of the above corrections
are already part of your SMS improvements patch for mainline (which I'll
approve shortly).  Anyway, I think it's best that these changes have a
chance to be tested on a range of platforms on mainline (or for another
maintainer to complain) before moving them to the gcc-4_0-branch, perhaps
on Wednesday?

My only minor suggestion is that stylistically I think its preferrable
to leave a blank line between the declarations at the start of a function
and the first line of imperative code in normalize_sched_times.

Thanks for this fix,

Roger
--