[Bug tree-optimization/16001] New: unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[Bug tree-optimization/16001] New: unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[rguenth at tat dot physik dot uni-tuebingen dot de]

g++ (GCC) 3.5-tree-ssa-lno 20040615 (merged 20040612)

rguenth@alwazn tmp $ ~/ix86/gccssa-lno/bin/g++ -O2 -funroll-loops -o
 tramp3d-v3.lno.lo tramp3d-v3.cpp -static -Dleafify=fooblah
tramp3d-v3.cpp:15627: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp:15636: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp:27426: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp:31114: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp:31390: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp:40288: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp:40296: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp:41213: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp:41399: warning: `fooblah' attribute directive ignored
tramp3d-v3.cpp: In member function `void
RefBlockController<T>::reallocateStorage(size_t, bool) [with T = Vector<3,
double, Full>]':
tramp3d-v3.cpp:10547: error: unable to find a register to spill in class
`AREG'
tramp3d-v3.cpp:10547: error: this is the insn:
(insn:HI 38 37 39 2 (parallel [
            (set (reg:SI 1 dx [93])
                (truncate:SI (lshiftrt:DI (mult:DI (zero_extend:DI (reg:SI
3 bx [91]))
                            (zero_extend:DI (reg:SI 1 dx [94])))
                        (const_int 32 [0x20]))))
            (clobber (scratch:SI))
            (clobber (reg:CC 17 flags))
        ]) 182 {*umulsi3_highpart_insn} (insn_list 37 (nil))
    (expr_list:REG_UNUSED (reg:CC 17 flags)
        (expr_list:REG_UNUSED (scratch:SI)
            (expr_list:REG_DEAD (reg:SI 3 bx [91])
                (expr_list:REG_DEAD (reg:SI 1 dx [94])
                    (expr_list:REG_UNUSED (reg:CC 17 flags)
                        (expr_list:REG_UNUSED (scratch:SI)
                            (expr_list:REG_EQUAL (truncate:SI (lshiftrt:DI
(mult:DI (zero_extend:DI (reg:SI 3 bx [91]))
                                            (const_double -1431655765
[0xaaaaaaab] 0 [0x0] 0 [0x0] 0 [0x0] 0 [0x0] 0 [0x0]))
                                        (const_int 32 [0x20])))
                                (nil)))))))))
tramp3d-v3.cpp:10547: confused by earlier errors, bailing out


Using -fpeel-loops instead of -funroll-loops makes the failure go away.

You can fetch tramp3d-v3.cpp from
http://www.tat.physik.uni-tuebingen.de/~rguenth/gcc/tramp3d-v3.cpp.gz

-- 
           Summary: unable to find a register to spill in class `AREG' (-O2
                    -funroll-loops)
           Product: gcc
           Version: lno
            Status: UNCONFIRMED
          Severity: critical
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at tat dot physik dot uni-tuebingen dot de
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |rtl-optimization
           Keywords|                            |ice-on-valid-code
   Target Milestone|---                         |lno


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2004-06-15 23:45 -------
Could you please provide a preprocessed source file?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[rakdver at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rakdver at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-06-16 07:47 -------
Confirmed.

I'm still reducing the testcase, however.
Down from 56k lines to 15k lines now. :-(
I'll post it when I have something more reasonable.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |reichelt at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-06-16 15:18 -------
Created an attachment (id=6540)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=6540&action=view)
still large testcase

I give up for now. I managed to get the testcase down to 6000 lines
and only one include file, namely <vector>.

I attached the whole preprocessed stuff (which is about 15000 lines).
In case somebody wants to factor out the include file, just
replace everything before the function "toss_cookies" by the
include file.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5? Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[belyshev at lubercy dot com]

------- Additional Comments From belyshev at lubercy dot com  2004-06-21 11:18 -------
'backported' to mainline. Introduced by:

2004-06-20  Zdenek Dvorak  <rakdver@atrey.karlin.mff.cuni.cz>

	* loop-invariant.c: New file.
	* Makefile.in (loop-invariant.o): New.
	* cfgloop.h (global_cost_for_size, init_set_costs,
	move_loop_invariants): Declare.
	* cfgloopanal.c (seq_cost, init_set_costs, global_cost_for_size): New
	functions.
	(avail_regs, res_regs, small_cost, pres_cost, spill_cost): New
	variables.
	* common.opt (floop-optimize2, fmove-loop-invariants): New options.
	* loop-init.c (loop_optimizer_init): Call init_set_costs.
	* passes.c (rest_of_handle_loop2): Call move_loop_invariants.
	(rest_of_compilation): Check flag_loop_optimize2.
	* toplev.c (process_options): Handle flag_loop_optimize2.
	* doc/invoke.texi (-floop-optimize2, -fmove-loop-invariants): Document.
	* doc/passes.texi (loop-invariant.c): Document.


... and then hidden (fixed?) by:

2004-06-21  Richard Henderson  <rth@redhat.com>

	* c-common.c (verify_sequence_points): Export.
	(c_expand_expr_stmt): Move to c-typeck.c.
	* c-common.h (c_expand_expr_stmt): Remove.
	(verify_sequence_points): Declare.
	* c-mudflap.c (mflang_flush_calls): Use c_finish_expr_stmt.
	* c-parse.in (for_init_stmt, stmt): Likewise.
	* c-tree.h (c_finish_expr_stmt): Declare.
	(c_tree_expr_nonnegative_p): Remove.
	* c-typeck.c (c_tree_expr_nonnegative_p): Remove.
	(build_conditional_expr, build_binary_op): Use tree_expr_nonnegative_p.
	(emit_side_effect_warnings): New.
	(c_finish_expr_stmt): Rename from c_expand_expr_stmt.  Use it.
	(c_finish_stmt_expr): Work without EXPR_STMT.  Handle eh regions.
	Use emit_side_effect_warnings.
	(push_cleanup): Copy STATEMENT_LIST_STMT_EXPR.
	* fold-const.c (tree_expr_nonnegative_p): Handle TARGET_EXPR.
	* gimplify.c (gimplify_modify_expr): Don't discard TARGET_EXPR
	with void initializer.
	(gimplify_target_expr): Handle void BIND_EXPR initializer.
	* tree-inline.c (estimate_num_insns_1): Fix type lookup for
	INIT_EXPR and MODIFY_EXPR.
	* objc/objc-act.c (build_module_descriptor): Use add_stmt
	instead of c_expand_expr_stmt.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
  GCC build triplet|i686-pc-linux-gnu           |
   GCC host triplet|i686-pc-linux-gnu           |
            Summary|unable to find a register to|[3.5? Regression] unable to
                   |spill in class `AREG' (-O2 -|find a register to spill in
                   |funroll-loops)              |class `AREG' (-O2 -funroll-
                   |                            |loops)


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5? Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[belyshev at lubercy dot com]

------- Additional Comments From belyshev at lubercy dot com  2004-06-21 18:22 -------
testcase in c:
-------------------------------------------------------------------------------
int reallocateStorage (int a)
{
  int j, k;
  k = foo ();
  j = a / 3;
  while (k)
    k ++;
  return j;
}
-------------------------------------------------------------------------------

Confirmed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
      Known to fail|                            |3.5.0
   Last reconfirmed|0000-00-00 00:00:00         |2004-06-21 18:22:53
               date|                            |
            Summary|[3.5? Regression] unable to |[3.5 Regression] unable to
                   |find a register to spill in |find a register to spill in
                   |class `AREG' (-O2 -funroll- |class `AREG' (-O2 -funroll-
                   |loops)                      |loops)
   Target Milestone|lno                         |3.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2004-06-21 19:18 -------
Patch:

http://gcc.gnu.org/ml/gcc-patches/2004-06/msg01676.html

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-06-21 19:24 -------
Subject: Bug 16001

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	lno-branch
Changes by:	rakdver@gcc.gnu.org	2004-06-21 19:24:14

Modified files:
	gcc            : ChangeLog.lno loop-iv.c 

Log message:
	PR rtl-optimization/16001
	* loop-iv.c (iv_number_of_iterations): Prevent copy propagation in
	niter_expr.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.lno.diff?cvsroot=gcc&only_with_tag=lno-branch&r1=1.1.2.196&r2=1.1.2.197
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/loop-iv.c.diff?cvsroot=gcc&only_with_tag=lno-branch&r1=1.1.4.11&r2=1.1.4.12



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-06-23 15:50 -------
Fixed on lno-branch by Zdenek's patch.

Zdenek, will your patch soon be seen in mainline, too?
(The problem is a regression on mainline.)

What about a testcase for the testsuite?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[rguenth at tat dot physik dot uni-tuebingen dot de]

------- Additional Comments From rguenth at tat dot physik dot uni-tuebingen dot de  2004-06-30 12:58 -------
It seems to be fixed on both lno and mainline as of today (Jun30) - but I cannot
find traces of the patch that went into lno on mainline.

Is the patch still needed on mainline?

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[giovannibajo at libero dot it]

------- Additional Comments From giovannibajo at libero dot it  2004-06-30 13:44 -------
The patch was never reviewed, this is why it didn't go into mainline yet, but 
it's still needed there I suppose, even if the bug cannot be reproduced with 
this testcase anymore.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |rakdver at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[rakdver at atrey dot karlin dot mff dot cuni dot cz]

------- Additional Comments From rakdver at atrey dot karlin dot mff dot cuni dot cz  2004-06-30 14:55 -------
Subject: Re:  [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

Hello,

> It seems to be fixed on both lno and mainline as of today (Jun30) - but I cannot
> find traces of the patch that went into lno on mainline.
> 
> Is the patch still needed on mainline?

yes, the patch is still needed, even if the problem was masked by some
other commit.

Zdenek


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-07-01 08:33 -------
I still (20040701) get an ICE on mainline with Serge's testcase as well
as with the following one (just compile with "gcc -O2 -funroll-loops"):

============================
int foo();

int bar(int i)
{
    int j = foo();
    i /= 3;
    while (j) ++j;
    return i;
}
============================


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[reichelt at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
           Keywords|                            |monitored
      Known to work|                            |3.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-07-11 02:52 -------
Subject: Bug 16001

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	rakdver@gcc.gnu.org	2004-07-11 02:52:54

Modified files:
	gcc            : ChangeLog loop-iv.c 

Log message:
	PR rtl-optimization/16001
	* loop-iv.c (iv_number_of_iterations): Prevent copy propagation in
	niter_expr.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.4457&r2=2.4458
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/loop-iv.c.diff?cvsroot=gcc&r1=2.11&r2=2.12



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001

[Bug rtl-optimization/16001] [3.5 Regression] unable to find a register to spill in class `AREG' (-O2 -funroll-loops)

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2004-07-11 02:53 -------
Fixed.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16001