* [PATCH, SMS] Fix calculation of issue_rate
@ 2011-05-19 10:33 Revital Eres
2011-05-19 10:44 ` Revital Eres
0 siblings, 1 reply; 2+ messages in thread
From: Revital Eres @ 2011-05-19 10:33 UTC (permalink / raw)
To: Ayal Zaks; +Cc: gcc-patches, Patch Tracking
Hello,
The issue rate is currently been set in SMS by calling
targetm.sched.issue_rate function if it is defined. For rs6000 the
issue_rate is 1 if !reload_completed && !flag_sched_pressure.
To bypass that, SMS sets reload_completed to 1 before calling
targetm.sched.issue_rate and restores it's original value after the call.
The problem is that the issue rate is changed again to 1 because of the
following chain of calls which occurs right after setting issue_rate
in targetm.sched.issue_rate ():
sms_schedule -> haifa_sched_init -> sched_init () ->
targetm.sched.issue_rate () (in haifa-sched.c:3474)
This time, when calling targetm.sched.issue_rate the issue_rate is
set to 1 as reload_completed contains it's original value (zero).
The attached patch tries to fix that.
Tested (bootstrap and regtest) on ppc64-redhat-linux.
OK for mainline?
Thanks,
Revital
Changelog:
* modulo-sched.c (sms_schedule): Fix stage_count calculation.
Index: modulo-sched.c
===================================================================
--- modulo-sched.c (revision 173786)
+++ modulo-sched.c (working copy)
@@ -924,6 +924,7 @@ sms_schedule (void)
basic_block condition_bb = NULL;
edge latch_edge;
gcov_type trip_count = 0;
+ int temp;
loop_optimizer_init (LOOPS_HAVE_PREHEADERS
| LOOPS_HAVE_RECORDED_EXITS);
@@ -933,22 +934,19 @@ sms_schedule (void)
return; /* There are no loops to schedule. */
}
+ temp = reload_completed;
+ reload_completed = 1;
/* Initialize issue_rate. */
if (targetm.sched.issue_rate)
- {
- int temp = reload_completed;
-
- reload_completed = 1;
- issue_rate = targetm.sched.issue_rate ();
- reload_completed = temp;
- }
+ issue_rate = targetm.sched.issue_rate ();
else
issue_rate = 1;
-
+
/* Initialize the scheduler. */
setup_sched_infos ();
haifa_sched_init ();
-
+ reload_completed = temp;
+
/* Allocate memory to hold the DDG array one entry for each loop.
We use loop->num as index into this array. */
g_arr = XCNEWVEC (ddg_ptr, number_of_loops ());
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH, SMS] Fix calculation of issue_rate
2011-05-19 10:33 [PATCH, SMS] Fix calculation of issue_rate Revital Eres
@ 2011-05-19 10:44 ` Revital Eres
0 siblings, 0 replies; 2+ messages in thread
From: Revital Eres @ 2011-05-19 10:44 UTC (permalink / raw)
To: Ayal Zaks; +Cc: gcc-patches
Hello,
>
> * modulo-sched.c (sms_schedule): Fix stage_count calculation.
>
Sorry, just noticed that the changelog entry is wrong, here
a correction:
* modulo-sched.c (sms_schedule): Fix issue_rate calculation.
Thanks,
Revital
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2011-05-19 7:00 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2011-05-19 10:33 [PATCH, SMS] Fix calculation of issue_rate Revital Eres
2011-05-19 10:44 ` Revital Eres
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).