diff mbox

Scheduler: Allow breaking dependencies by modifying patterns

Message ID 505B937B.3080506@codesourcery.com
State New
Headers show

Commit Message

Bernd Schmidt Sept. 20, 2012, 10:06 p.m. UTC 
On 09/20/2012 07:42 PM, H.J. Lu wrote:
> On Fri, Aug 3, 2012 at 5:05 AM, Bernd Schmidt <bernds@codesourcery.com> wrote:
>> This patch allows us to change
>>
>> rn++
>> rm=[rn]
>>
>> into
>>
>> rm=[rn + 4]
>> rn++
>>
> This caused:
> 
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54645

The jury is still out on that, but it did cause PR54643. This seems to
be some kind of interaction with the new SCHED_PRESSURE_MODEL code. For
now it seems best to just disable the new optimization if scheduling for
pressure. It will still trigger during sched2.

I've bootstrapped the following on x86_64-linux and verified that the
arm build gets further (it now fails in libbacktrace). Committed.


Bernd
diff mbox

Patch

Index: gcc/ChangeLog
===================================================================
--- gcc/ChangeLog	(revision 191592)
+++ gcc/ChangeLog	(working copy)
@@ -1,3 +1,9 @@ 
+2012-09-20  Bernd Schmidt  <bernds@codesourcery.com>
+
+	PR bootstrap/54643
+	* haifa-sched.c (schedule_block): Skip find_modifiable_mems if using
+	SCHED_PRESSURE_MODEL.
+
 2012-09-20  Joern Rennecke  <joern.rennecke@embecosm.com>
 
 	* rtlanal.c (dead_or_set_regno_p): Fix COND_EXEC handling.
Index: gcc/haifa-sched.c
===================================================================
--- gcc/haifa-sched.c	(revision 191582)
+++ gcc/haifa-sched.c	(working copy)
@@ -5854,7 +5854,8 @@  schedule_block (basic_block *target_bb)
   rtx head = NEXT_INSN (prev_head);
   rtx tail = PREV_INSN (next_tail);
 
-  if ((current_sched_info->flags & DONT_BREAK_DEPENDENCIES) == 0)
+  if ((current_sched_info->flags & DONT_BREAK_DEPENDENCIES) == 0
+      && sched_pressure != SCHED_PRESSURE_MODEL)
     find_modifiable_mems (head, tail);
 
   /* We used to have code to avoid getting parameters moved from hard