[PATCH resubmit] SMS improvements 2/2

[Mostafa Hagog <MUSTAFA at il dot ibm dot com>]

As a follow up to http://gcc.gnu.org/ml/gcc-patches/2005-03/msg01892.html I
have addressed the comment there in the following patch.  the regression
and bootstrap passed for the previous patch on i686-pc-linux-gnu and
powerpc-apple-darwin. This patch has only comment changes and part of the
previous patch was already committed as part of a bug fix in SMS. Bootstrap
of the new patch on both targets to make sure I didn't miss something.

Mostafa.

2005-03-31 Mostafa Hagog <mustafa@il.ibm.com>

      * cfg.c (clear_bb_flags): Don't clear BB_DISABLE_SCHEDULE.
      * modulo-sched.c (undo_replace_buff_elem): New structure.
      (kernel_number_of_cycles, ps_unschedule_node,
      undo_generate_reg_moves,free_undo_replace_buff,
      undo_permute_partial_schedule,      loop_single_full_bb_p,
      SIMPLE_SMS_LOOP_P, loop_canon_p, canon_loop,
      build_loops_structure, get_sched_window): New.
      (generate_reg_moves): Return undo_replace_buff_elem and other
      fixes.
      (generate_prolog_epilog): Remove old loop versioning.
      (sms_schedule): Use loop information and loop_version.
      (sms_schedule_by_order): Split part of it to get_sched_window.
      * passes.c (rest_of_handle_sms): call cfg_layout_initialize
      cfg_layout_finalize and free_dominance_info before/after SMS.

(See attached file: sms_main-8.patch)

Attachment: sms_main-8.patch
Description: Binary data