[Bug rtl-optimization/46829] New: ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[Bug rtl-optimization/46829] New: ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

           Summary: ICE: in spill_failure, at reload1.c:2105 with
                    -fschedule-insns -fsched-pressure and variadic
                    function
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: zsojka@seznam.cz
              Host: x86_64-pc-linux-gnu
            Target: x86_64-pc-linux-gnu


Created attachment 22663
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22663
reduced testcase

Compiler flags:
gcc -O[123] -fschedule-insns -fsched-pressure

Compiler output:
$ gcc -O -fschedule-insns -fsched-pressure pr46829.c 
pr46829.c: In function 'foo':
pr46829.c:15:1: error: unable to find a register to spill in class 'AREG'
pr46829.c:15:1: error: this is the insn:
(insn 16 21 18 4 (parallel [
            (set (reg:SI 37 r8)
                (div:SI (reg:SI 37 r8 [orig:59 D.2694 ] [59])
                    (reg:SI 38 r9 [orig:60 D.2695 ] [60])))
            (set (reg:SI 1 dx [69])
                (mod:SI (reg:SI 37 r8 [orig:59 D.2694 ] [59])
                    (reg:SI 38 r9 [orig:60 D.2695 ] [60])))
            (clobber (reg:CC 17 flags))
        ]) pr46829.c:14 353 {*divmodsi4}
     (expr_list:REG_DEAD (reg:SI 37 r8 [orig:59 D.2694 ] [59])
        (expr_list:REG_UNUSED (reg:SI 1 dx [69])
            (expr_list:REG_UNUSED (reg:CC 17 flags)
                (nil)))))
pr46829.c:15:1: internal compiler error: in spill_failure, at reload1.c:2105
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

Tested revisions:
r167471 - crash
4.5.1 - crash

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2010.12.06 21:53:26
     Ever Confirmed|0                           |1

--- Comment #1 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-06 21:53:26 UTC ---
Confirmed with revision 167517.

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

--- Comment #2 from Zdenek Sojka <zsojka at seznam dot cz> 2010-12-06 22:17:44 UTC ---
This ICE is quite unlucky, it happened in 3rd package I tried to compile - now
I have compiled 70 packages since then, and this is still the only occurence.

(don't get me wrong, I really admire the work done in scheduling! -and in fact,
in all of gcc)

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[changpeng.fang at amd dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

Changpeng Fang <changpeng.fang at amd dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |changpeng.fang at amd dot
                   |                            |com

--- Comment #3 from Changpeng Fang <changpeng.fang at amd dot com> 2010-12-06 22:41:29 UTC ---
Actually, I have opened bug 46793 last week, and it seems to be the same bug.
Bug 46793 has been marked as WONTFIX.

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-06 22:52:04 UTC ---
(In reply to comment #3)
> Actually, I have opened bug 46793 last week, and it seems to be the same bug.
> Bug 46793 has been marked as WONTFIX.

No, it isn't. -fschedule-insns is bad for x86. "-fschedule-insns
-fsched-pressure"
should work.

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-12-07 11:21:45 UTC ---
It should work for x86_64, not necessarily i?86.

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

--- Comment #6 from Zdenek Sojka <zsojka at seznam dot cz> 2010-12-07 17:34:01 UTC ---
Created attachment 22674
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22674
different testcase

A bit different testcase

$ gcc -O -fschedule-insns -fsched-pressure pr46829-2.c 
pr46829-2.c: In function 'foo':
pr46829-2.c:10:1: error: unable to find a register to spill in class 'AREG'
pr46829-2.c:10:1: error: this is the insn:
(insn 8 13 9 2 (parallel [
            (set (reg:SI 37 r8)
                (div:SI (reg:SI 2 cx [orig:59 b.0 ] [59])
                    (reg:SI 38 r9 [orig:60 a.1 ] [60])))
            (set (reg:SI 1 dx [66])
                (mod:SI (reg:SI 2 cx [orig:59 b.0 ] [59])
                    (reg:SI 38 r9 [orig:60 a.1 ] [60])))
            (clobber (reg:CC 17 flags))
        ]) pr46829-2.c:8 353 {*divmodsi4}
     (expr_list:REG_UNUSED (reg:SI 1 dx [66])
        (expr_list:REG_UNUSED (reg:CC 17 flags)
            (nil))))
pr46829-2.c:10:1: internal compiler error: in spill_failure, at reload1.c:2105
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[changpeng.fang at amd dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

--- Comment #7 from Changpeng Fang <changpeng.fang at amd dot com> 2010-12-07 22:42:26 UTC ---
If you compile with dispatch scheduling enabled (together with -march=bdver1),
both test cases attached PASS.

gcc -O2 -fschedule-insns -fsched-pressure -mdispatch-scheduler -march=bdver1

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

--- Comment #8 from H.J. Lu <hjl.tools at gmail dot com> 2010-12-07 23:54:42 UTC ---
(In reply to comment #7)
> If you compile with dispatch scheduling enabled (together with -march=bdver1),
> both test cases attached PASS.
> 
> gcc -O2 -fschedule-insns -fsched-pressure -mdispatch-scheduler -march=bdver1

You can turn it on for -march=bdver1.

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[changpeng.fang at amd dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

Changpeng Fang <changpeng.fang at amd dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vmakarov at redhat dot com

--- Comment #9 from Changpeng Fang <changpeng.fang at amd dot com> 2010-12-09 19:16:57 UTC ---
(In reply to comment #5)
> It should work for x86_64, not necessarily i?86.

Do you mean -fsched-pressure should be able to solve the problem completely
for x86-64?

Vladimir: Do you have any idea which direction to go in order to solve this
problem?

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[vmakarov at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

--- Comment #10 from Vladimir Makarov <vmakarov at redhat dot com> 2010-12-10 15:45:33 UTC ---
(In reply to comment #9)
> (In reply to comment #5)
> > It should work for x86_64, not necessarily i?86.
> 
> Do you mean -fsched-pressure should be able to solve the problem completely
> for x86-64?
> 
> Vladimir: Do you have any idea which direction to go in order to solve this
> problem?

Introducing of -fsched-pressure just decreased probability of the bug when 1st
insn scheduling is used.  The patch introducing -fsched-pressure contained some
code in reload to decrease the probability even more.

Unfortunately, it did not eliminated it fully.  This bug can not be fixed in
scheduler (or the solution, like not moving through insn referring for a hard
register, will be too conservative especially for x86_64 and still will not fix
it for x86) because the scheduler can not see all info handled by reload.

IMHO, the right fix should be possibility to split live ranges for explicitly
mentioned hard register.  May be Jeff Law's current work will provide such
feature.

In any case, implementation of live range splitting in reload is too big and
complicated job even for stage #1.  There is no way to implement it at stage
#3.  It is also very unreasonable thing to do because any change in reload is
usually very bug prone.

I am sorry, but I don't see that it can be fixed for gcc4.6 fox x86/x86-64. 
Although it might be fixed for gcc4.7.

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
                URL|                            |http://gcc.gnu.org/ml/gcc-p
                   |                            |atches/2012-08/msg01202.htm
                   |                            |l
         AssignedTo|unassigned at gcc dot       |ubizjak at gmail dot com
                   |gnu.org                     |

--- Comment #11 from Uros Bizjak <ubizjak at gmail dot com> 2012-08-17 18:37:44 UTC ---
Patch at [1].

[1] http://gcc.gnu.org/ml/gcc-patches/2012-08/msg01202.html

[Bug rtl-optimization/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[uros at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

--- Comment #12 from uros at gcc dot gnu.org 2012-09-01 14:29:51 UTC ---
Author: uros
Date: Sat Sep  1 14:29:45 2012
New Revision: 190847

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=190847
Log:
    PR target/46829
    PR target/46843
    * config/i386/i386.c (ix86_legitimate_combined_insn): New function.
    (TARGET_LEGITIMATE_COMBINED_INSN): New macro.

testsuite/ChangeLog:

    PR target/46829
    PR target/46843
    * gcc.target/i386/pr46829.c: New test.
    * gcc.target/i386/pr46843.c: Ditto.


Added:
    trunk/gcc/testsuite/gcc.target/i386/pr46829.c
    trunk/gcc/testsuite/gcc.target/i386/pr46843.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/i386/i386.c
    trunk/gcc/testsuite/ChangeLog

[Bug target/46829] ICE: in spill_failure, at reload1.c:2105 with -fschedule-insns -fsched-pressure and variadic function

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46829

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
          Component|rtl-optimization            |target
         Resolution|                            |FIXED
   Target Milestone|---                         |4.8.0

--- Comment #13 from Uros Bizjak <ubizjak at gmail dot com> 2012-09-05 08:20:14 UTC ---
Fixed for 4.8.