[Bug rtl-optimization/15521] New: sched1 extends life of hard registers on SMALL_REGISTER_CLASSES targets when exceptions are enabled

[Bug rtl-optimization/15521] New: sched1 extends life of hard registers on SMALL_REGISTER_CLASSES targets when exceptions are enabled

[amylaar at gcc dot gnu dot org]

The current mailine fails to build libstdc++-v3 for sh-elf.  We get
a reload failure compiling locale-inst.cc for -m4-single-only .
The first scheduling pass moves the instruction that moves the call
value from a hard to a pseudo register away from the call, which extends
the lifetime of r0 across an instruction with indexed addressing.
The problem is that the call is in a different scheduling region, as
exception handling causes it to end a basic block.
Thus, sched_analyze doesn't see the CALL_INSN during the current region
scheduling, in_post_call_group_p does not get set, and thus the return
value copy instruction doesn't get the SCHED_GROUP_P bit set.

A patch is here:
http://gcc.gnu.org/ml/gcc-patches/2004-05/msg00857.html

-- 
           Summary: sched1 extends life of hard registers on
                    SMALL_REGISTER_CLASSES targets when exceptions are
                    enabled
           Product: gcc
           Version: 3.5.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code, build, patch, EH
          Severity: critical
          Priority: P2
         Component: rtl-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: amylaar at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: any
  GCC host triplet: any
GCC target triplet: sh-elf and other SMALL_REGISTER_CLASSES targets


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15521

[Bug rtl-optimization/15521] [3.5 Regression] sched1 extends life of hard registers on SMALL_REGISTER_CLASSES targets when exceptions are enabled

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-05-18 19:47 -------
Confirmed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2004-05-18 19:47:36
               date|                            |
            Summary|sched1 extends life of hard |[3.5 Regression] sched1
                   |registers on                |extends life of hard
                   |SMALL_REGISTER_CLASSES      |registers on
                   |targets when exceptions are |SMALL_REGISTER_CLASSES
                   |enabled                     |targets when exceptions are
                   |                            |enabled
   Target Milestone|---                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15521

[Bug rtl-optimization/15521] [3.5 Regression] sched1 extends life of hard registers on SMALL_REGISTER_CLASSES targets when exceptions are enabled

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2004-05-27 16:36 -------
Blocks merge of a lot of patches.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P2                          |P1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15521

[Bug rtl-optimization/15521] [3.5 Regression] sched1 extends life of hard registers on SMALL_REGISTER_CLASSES targets when exceptions are enabled

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2004-06-04 16:08 -------
The control flow manipulation also can separate the call
from the return value copy, a patch is here:
http://gcc.gnu.org/ml/gcc-patches/2004-06/msg00229.html

A build also failed due to an unrelated bug in real.c, a patch is
here: http://gcc.gnu.org/ml/gcc-patches/2004-06/msg00227.html

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15521

[Bug rtl-optimization/15521] [3.5 Regression] sched1 extends life of hard registers on SMALL_REGISTER_CLASSES targets when exceptions are enabled

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-06-09 12:48 -------
Subject: Bug 15521

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	amylaar@gcc.gnu.org	2004-06-09 12:48:22

Modified files:
	gcc            : ChangeLog sched-int.h sched-deps.c 

Log message:
	PR rtl-optimization/15521:
	* sched-int.h (in_post_call_group_p): Change type to enum.
	* sched-deps.c (sched_analyze_insn):
	(sched_analyze): When in_post_call_group_p is post_call_initial,
	don't add a dependency, but still set SCHED_GROUP_P and CANT_MOVE,
	and also reset in_post_call_group_p to post_call.
	(sched_analyze): When the previous basic block ended in a CALL_INSN,
	initialize in_post_call_group_p as post_call_initial.
	(init_deps): initialize in_post_call_group_p to not_post_call.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.3886&r2=2.3887
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/sched-int.h.diff?cvsroot=gcc&r1=1.33&r2=1.34
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/sched-deps.c.diff?cvsroot=gcc&r1=1.72&r2=1.73



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15521

[Bug rtl-optimization/15521] [3.5 Regression] sched1 extends life of hard registers on SMALL_REGISTER_CLASSES targets when exceptions are enabled

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-06-09 14:09 -------
Fixed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15521