[PATCH 4/8] Disable max_issue when scheduling for register pressure

[PATCH 4/8] Disable max_issue when scheduling for register pressure

[Maxim Kuvyrkov]

[-- Attachment #1: Type: text/plain, Size: 634 bytes --]

Hi,

This patch disables max_issue-based lookahead multipass scheduling when register-pressure heuristics are active.  The two approaches tend to undo each others decisions and don't play well together.

Currently this patch is a no-op, since no target uses both max_issue and register-pressure scheduling.  This will change soon with upcoming patches that will enable max_issue during 2nd scheduling pass for ARM and AArch64.

Bootstrapped on x86_64-linux-gnu.  Regression testing is in progress.  OK to commit if no regressions?

[ChangeLog is part of the git patch]

Thank you,

--
Maxim Kuvyrkov
www.linaro.org



[-- Attachment #2: 0004-Disable-max_issue-when-scheduling-for-register-press.patch --]
[-- Type: application/octet-stream, Size: 1071 bytes --]

From caae8fd31b92ea63d91604df4d71a7866303ec2a Mon Sep 17 00:00:00 2001
From: Maxim Kuvyrkov <maxim.kuvyrkov@linaro.org>
Date: Tue, 21 Oct 2014 00:25:11 +0100
Subject: [PATCH 4/8] Disable max_issue when scheduling for register pressure

	* haifa-sched.c (sched_init): Disable max_issue when scheduling for
	register pressure.
---
 gcc/haifa-sched.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/gcc/haifa-sched.c b/gcc/haifa-sched.c
index 4f9648c..26d9e29 100644
--- a/gcc/haifa-sched.c
+++ b/gcc/haifa-sched.c
@@ -6867,7 +6867,10 @@ sched_init (void)
   else
     issue_rate = 1;
 
-  if (targetm.sched.first_cycle_multipass_dfa_lookahead)
+  if (targetm.sched.first_cycle_multipass_dfa_lookahead
+      /* Don't use max_issue with reg_pressure scheduling.  Multipass
+	 scheduling and reg_pressure scheduling undo each other's decisions.  */
+      && sched_pressure == SCHED_PRESSURE_NONE)
     dfa_lookahead = targetm.sched.first_cycle_multipass_dfa_lookahead ();
   else
     dfa_lookahead = 0;
-- 
1.7.9.5

Re: [PATCH 4/8] Disable max_issue when scheduling for register pressure

[Vladimir Makarov]

On 10/20/2014 11:21 PM, Maxim Kuvyrkov wrote:
> Hi,
>
> This patch disables max_issue-based lookahead multipass scheduling when register-pressure heuristics are active.  The two approaches tend to undo each others decisions and don't play well together.
Yes, I am agree.
> Currently this patch is a no-op, since no target uses both max_issue and register-pressure scheduling.  This will change soon with upcoming patches that will enable max_issue during 2nd scheduling pass for ARM and AArch64.
>
> Bootstrapped on x86_64-linux-gnu.  Regression testing is in progress.  OK to commit if no regressions?
>
The patch is ok.  Thanks, Maxim.