[PATCH] Don't call dbr_schedule when -O0 -fdelayed-branch (PR rtl-optimization/38740)

[Jakub Jelinek <jakub at redhat dot com>]

Hi!

The testcase in the PR is miscompiled when compiled with weirdo option
combination - -O0 -fdelayed-branch.  The problem is that df_analyze
is only called after RA if optimize > 0 and dbr_schedule relies on the
dataflow info.  While we could df_analyze at the end of the IRA pass
if (optimize || flag_delayed_branch) instead of just if (optimize),
I don't think it is worth supporting such weirdo option combination.
This patch instead doesn't dbr_schedule for -O0.
sh.c, the only other caller of dbr_schedule, already calls
it only if optimize > 0 && flag_delayed_branch.

Ok for trunk?  No testcases that I see use -fdelayed-branch
and at -O0 flag_delayed_branch defaults to 0, so this patch shouldn't affect
bootstrap nor regtest on any target.

2009-01-28  Jakub Jelinek  <jakub@redhat.com>

	PR rtl-optimization/38740
	* reorg.c (gate_handle_delay_slots): Avoid dbr scheduling
	if !optimize.
	* config/mips/mips.c (mips_reorg): Likewise.

--- gcc/reorg.c.jj	2009-01-13 18:32:17.000000000 +0100
+++ gcc/reorg.c	2009-01-28 16:33:55.000000000 +0100
@@ -4046,7 +4046,8 @@ static bool
 gate_handle_delay_slots (void)
 {
 #ifdef DELAY_SLOTS
-  return flag_delayed_branch && !crtl->dbr_scheduled_p;
+  /* At -O0 dataflow info isn't updated after RA.  */
+  return optimize > 0 && flag_delayed_branch && !crtl->dbr_scheduled_p;
 #else
   return 0;
 #endif
--- gcc/config/mips/mips.c.jj	2009-01-13 18:32:17.000000000 +0100
+++ gcc/config/mips/mips.c	2009-01-28 16:32:47.000000000 +0100
@@ -13296,7 +13296,7 @@ mips_reorg (void)
   mips16_lay_out_constants ();
   if (mips_r10k_cache_barrier != R10K_CACHE_BARRIER_NONE)
     r10k_insert_cache_barriers ();
-  if (flag_delayed_branch)
+  if (optimize > 0 && flag_delayed_branch)
     dbr_schedule (get_insns ());
   mips_reorg_process_insns ();
   if (!TARGET_MIPS16

	Jakub