[Bug rtl-optimization/54294] New: [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[Bug rtl-optimization/54294] New: [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

             Bug #: 54294
           Summary: [4.8 Regression]: [alpha] Bootstrap comparison failure
                    due to fwprop handling of debug insns
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Keywords: build
          Severity: normal
          Priority: P3
         Component: rtl-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ubizjak@gmail.com
            Target: alphaev68-unknown-linux-gnu


Alpha bootstrap is currently broken due to stage2/3 comparison failure.

The problem is in handling of debug instructions in fwprop1 pass.

Please compare following non-debug compilation, where we start from
_.ii.156r.cse1 pass:

(note 3642 3637 3643 423 [bb 423] NOTE_INSN_BASIC_BLOCK)
(insn 3643 3642 3644 423 (set (reg:DI 1212 [ ei ])
        (sign_extend:DI (subreg:SI (reg:DI 1212 [ ei ]) 0))) 2 {*extendsidi2_1}
     (nil))
(insn 3644 3643 3728 423 (set (reg:DI 1263 [ ivtmp.371 ])
        (sign_extend:DI (subreg:SI (reg:DI 1212 [ ei ]) 0))) 2 {*extendsidi2_1}
     (nil))

And in _.ii.157.fwprop1, operands got propagated to:

(note 3642 3637 3643 423 [bb 423] NOTE_INSN_BASIC_BLOCK)
(insn 3643 3642 3644 423 (set (reg:DI 1212 [ ei ])
        (sign_extend:DI (reg:SI 2385))) 2 {*extendsidi2_1}
     (nil))
(insn 3644 3643 3728 423 (set (reg:DI 1263 [ ivtmp.371 ])
        (sign_extend:DI (reg:SI 2385))) 2 {*extendsidi2_1}
     (nil))


The difference in debug compilation can be shown starting again from
_.ii.156r.cse1 pass:

(note 4153 4148 4154 423 [bb 423] NOTE_INSN_BASIC_BLOCK)
(insn 4154 4153 4155 423 (set (reg:DI 1212 [ ei ])
        (sign_extend:DI (subreg:SI (reg:DI 1212 [ ei ]) 0))) 2 {*extendsidi2_1}
     (nil))
(debug_insn 4155 4154 4157 423 (var_location:SI ei (subreg:SI (reg:DI 1212 [ ei
]) 0)) -1
     (nil))
(debug_insn 4157 4155 4158 423 (var_location:BLK D#9 (clobber (const_int 0
[0]))) ../../gcc-svn/trunk/gcc/cprop.c:1519 -1
     (nil))
(debug_insn 4158 4157 4159 423 (var_location:DI p (debug_implicit_ptr:DI e))
../../gcc-svn/trunk/gcc/cprop.c:1519 -1
     (nil))
(debug_insn 4159 4158 4160 423 (var_location:SI ei (subreg:SI (reg:DI 1212 [ ei
]) 0)) -1
     (nil))
(debug_insn 4160 4159 4162 423 (var_location:DI ei$8 (debug_expr:DI D#34)) -1
     (nil))
(debug_insn 4162 4160 4163 423 (var_location:BLK D#7 (clobber (const_int 0
[0]))) ../../gcc-svn/trunk/gcc/basic-block.h:636 -1
     (nil))
(debug_insn 4163 4162 4164 423 (var_location:DI i$container (debug_expr:DI
D#34)) -1
     (nil))
(debug_insn 4164 4163 4165 423 (var_location:DI vec_ (clobber (const_int 0
[0]))) ../../gcc-svn/trunk/gcc/basic-block.h:586 -1
     (nil))
(insn 4165 4164 4249 423 (set (reg:DI 1263 [ ivtmp.371 ])
        (sign_extend:DI (subreg:SI (reg:DI 1212 [ ei ]) 0))) 2 {*extendsidi2_1}
     (nil))

where _.ii.157.fwprop1 pass propagates only operands before the clobber,
resulting in:

(note 4153 4148 4154 423 [bb 423] NOTE_INSN_BASIC_BLOCK)
(insn 4154 4153 4155 423 (set (reg:DI 1212 [ ei ])
        (sign_extend:DI (reg:SI 2385))) 2 {*extendsidi2_1}
     (nil))
(debug_insn 4155 4154 4157 423 (var_location:SI ei (reg:SI 2385)) -1
     (nil))
(debug_insn 4157 4155 4158 423 (var_location:BLK D#9 (clobber (const_int 0
[0]))) ../../gcc-svn/trunk/gcc/cprop.c:1519 -1
     (nil))
(debug_insn 4158 4157 4159 423 (var_location:DI p (debug_implicit_ptr:DI e))
../../gcc-svn/trunk/gcc/cprop.c:1519 -1
     (nil))
(debug_insn 4159 4158 4160 423 (var_location:SI ei (subreg:SI (reg:DI 1212 [ ei
]) 0)) -1
     (nil))
(debug_insn 4160 4159 4162 423 (var_location:DI ei$8 (debug_expr:DI D#34)) -1
     (nil))
(debug_insn 4162 4160 4163 423 (var_location:BLK D#7 (clobber (const_int 0
[0]))) ../../gcc-svn/trunk/gcc/basic-block.h:636 -1
     (nil))
(debug_insn 4163 4162 4164 423 (var_location:DI i$container (debug_expr:DI
D#34)) -1
     (nil))
(debug_insn 4164 4163 4165 423 (var_location:DI vec_ (clobber (const_int 0
[0]))) ../../gcc-svn/trunk/gcc/basic-block.h:586 -1
     (nil))
(insn 4165 4164 4249 423 (set (reg:DI 1263 [ ivtmp.371 ])
        (sign_extend:DI (subreg:SI (reg:DI 1212 [ ei ]) 0))) 2 {*extendsidi2_1}
     (nil))

This introduces real difference in (insn 4165) of the debug compilation vs
(insn 3644) of the non-debug compilation. Things go down from there, the final
difference is (non-debug):

(insn:TI 3643 2534 3644 (set (reg:DI 2 $2 [orig:1212 ei ] [1212])
        (sign_extend:DI (plus:SI (reg:SI 4 $4 [orig:1263 ivtmp.371 ] [1263])
                (const_int 1 [0x1])))) 4 {*addsi_se}
     (expr_list:REG_DEAD (reg:SI 4 $4 [orig:1263 ivtmp.371 ] [1263])
        (nil)))
(insn:TI 3644 3643 6425 (set (reg:DI 4 $4 [orig:1263 ivtmp.371 ] [1263])
        (reg:DI 2 $2 [orig:1212 ei ] [1212])) 236 {*movdi}
     (nil))

versus debug compilation:

(note 8075 2865 8076 (var_location i (debug_implicit_ptr:DI ei))
NOTE_INSN_VAR_LOCATION)
(note 8076 8075 4154 (var_location i$container (reg/v/f:DI 11 $11 [orig:754 bb
] [754])) NOTE_INSN_VAR_LOCATION)
(insn:TI 4154 8076 8077 (set (reg:DI 2 $2 [orig:1212 ei ] [1212])
        (sign_extend:DI (plus:SI (reg:SI 4 $4 [orig:1263 ivtmp.371 ] [1263])
                (const_int 1 [0x1])))) 4 {*addsi_se}
     (expr_list:REG_DEAD (reg:SI 4 $4 [orig:1263 ivtmp.371 ] [1263])
        (nil)))
(note 8077 4154 8078 (var_location p (debug_implicit_ptr:DI e))
NOTE_INSN_VAR_LOCATION)
(note 8078 8077 4165 (var_location i$container (reg/v/f:DI 11 $11 [orig:754 bb
] [754])) NOTE_INSN_VAR_LOCATION)
(insn:TI 4165 8078 6973 (set (reg:DI 4 $4 [orig:1263 ivtmp.371 ] [1263])
        (sign_extend:DI (reg:SI 2 $2 [orig:1212 ei ] [1212]))) 2
{*extendsidi2_1}
     (nil))

and finally:

+       .loc 1 1520 0
        ldq $2,8($2)
        ldl $2,48($2)
        and $2,64,$2
        bne $2,$L829
+$LVL612:
        addl $4,1,$2
-       mov $2,$4
+       addl $31,$2,$4
        br $31,$L421
+$LVL613:
 $L829:

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2012-08-16 23:23:39 UTC ---
Created attachment 28040
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28040
Preprocessed source of the object file that triggers comparison failure

~/gcc-build/stage1-gcc/xgcc -B ~/gcc-build/stage1-gcc -O2 -fno-exceptions
-fno-rtti -fno-common -fcompare-debug cprop.ii
xgcc: error: cprop.ii: -fcompare-debug failure (length)

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

Uros Bizjak <ubizjak at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |aoliva at gcc dot gnu.org,
                   |                            |steven at gcc dot gnu.org

--- Comment #2 from Uros Bizjak <ubizjak at gmail dot com> 2012-08-16 23:25:30 UTC ---
Adding come CCs that might be interested in this issue.

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

--- Comment #3 from Uros Bizjak <ubizjak at gmail dot com> 2012-08-17 08:10:07 UTC ---
(In reply to comment #1)
> Created attachment 28040 [details]
> Preprocessed source of the object file that triggers comparison failure
> 
> ~/gcc-build/stage1-gcc/xgcc -B ~/gcc-build/stage1-gcc -O2 -fno-exceptions
> -fno-rtti -fno-common -fcompare-debug cprop.ii
> xgcc: error: cprop.ii: -fcompare-debug failure (length)

Adding -fno-forward-propagate to the above flags fixes compare-debug failure.

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2012-08-17
                 CC|                            |jakub at gcc dot gnu.org
         AssignedTo|unassigned at gcc dot       |jakub at gcc dot gnu.org
                   |gnu.org                     |
     Ever Confirmed|0                           |1

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-08-17 08:39:24 UTC ---
Looking into it.

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bonzini at gnu dot org
   Target Milestone|---                         |4.8.0

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-08-17 10:29:20 UTC ---
Created attachment 28041
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28041
gcc48-pr54294.patch

Seems my other PR42728 patch fixes the -fcompare-debug failure and is IMHO
desirable in any case.  What is going on here is that we have:
(insn 14254 14253 512 423 (set (reg:DI 1196 [ ei ])
        (sign_extend:DI (subreg:SI (reg:DI 1196 [ ei ]) 0))) 2 {*extendsidi2_1}
     (nil))
insn where (subreg:SI (reg:DI 1196 [ ei ]) 0) is replaced by (reg:SI 2915). 
But fwprop is df_set_flags (DF_DEFER_INSN_RESCAN);.  When replacing the next
insn,
which without -g turns to be another set, but with -g there are several debug
insns in between and next insn is the first of them, we take a shortcut.  As
rescan is deferred, we see both 1196 and 2915 regs as "uses" of def_insn.  The
shortcut performs a rtx_equal_p check on DF_REF_REG (use) which fails though,
as DF_REF_REG (use) is the subreg, thus we replace the subreg in the
immediately following insn.  But when trying to replace it in following debug
insns and with -g in the first following non-debug insn, it is not NEXT_INSN,
shortcut is not taken and we do both the rtx_equal_p check (fine, DF_REF_REG is
again a subreg), but also do use_killed_between including def_insn, and that
calls local_ref_killed_between_p, which compares DF_REF_REGNO instead of using
rtx_equal_p and therefore says that the (already non-existing) use in def_insn
is killed by the set in the def_insn.

So, IMHO we should possibly replace the rtx_equal_p checks with simple
DF_REF_REGNO comparisons instead, which will be more conservatively correct,
but we risk code quality regressions.

So, perhaps another thing to do could be that when walking uses of def_insn in
all_uses_available_at, we'd double check if the reg in question is really still
used in the insn pattern (perhaps conditionally on whether the insn is still
marked for rescanning or not).

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

--- Comment #6 from Uros Bizjak <ubizjak at gmail dot com> 2012-08-17 11:14:45 UTC ---
(In reply to comment #5)
> Created attachment 28041 [details]
> gcc48-pr54294.patch

This patch fixes the debug-compare failure with testcase on alpha. I am running
bootstrap now.

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

--- Comment #7 from Uros Bizjak <ubizjak at gmail dot com> 2012-08-17 14:25:52 UTC ---
(In reply to comment #6)
> (In reply to comment #5)
> > Created attachment 28041 [details]
> > gcc48-pr54294.patch
> 
> This patch fixes the debug-compare failure with testcase on alpha. I am running
> bootstrap now.

Bootstrap finished OK, all interesting regression test passed on
alphaev68-unknown-linux-gnu.

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

--- Comment #8 from Uros Bizjak <ubizjak at gmail dot com> 2012-08-17 19:50:10 UTC ---
(In reply to comment #7)

> Bootstrap finished OK, all interesting regression test passed on
> alphaev68-unknown-linux-gnu.

Regression test is also ok [1].

[1] http://gcc.gnu.org/ml/gcc-testresults/2012-08/msg01654.html

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-08-20 18:56:58 UTC ---
Author: jakub
Date: Mon Aug 20 18:56:49 2012
New Revision: 190541

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=190541
Log:
    PR rtl-optimization/54294
    * fwprop.c (all_uses_available_at): Ignore debug insns in between
    def_insn and target_insn when checking whether the shortcut is
    possible.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/fwprop.c

[Bug rtl-optimization/54294] [4.8 Regression]: [alpha] Bootstrap comparison failure due to fwprop handling of debug insns

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54294

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-08-20 18:58:24 UTC ---
Fixed.