[Bug debug/104263] New: [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions

[Bug debug/104263] New: [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions

[asolokha at gmx dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

            Bug ID: 104263
           Summary: [10/11/12 Regression] '-fcompare-debug' failure
                    (length) w/ -O2 -fnon-call-exceptions
                    -fno-inline-small-functions
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: compare-debug-failure
          Severity: normal
          Priority: P3
         Component: debug
          Assignee: unassigned at gcc dot gnu.org
          Reporter: asolokha at gmx dot com
  Target Milestone: ---

gcc 12.0.1 20220123 snapshot (g:2da90ad39bf8fa9ee287e040d1f4411cb7a2e7ed) fails
-fcompare-debug check when compiling the following testcase w/ -O2
-fnon-call-exceptions -fno-inline-small-functions:

int n;

int
bar (void)
{
  int a;

  n = 0;
  a = 0;

  return n;
}

__attribute__ ((pure, returns_twice)) int
foo (void)
{
  n = bar () + 1;
  foo ();

  return 0;
}

% gcc-12.0.1 -O2 -fcompare-debug -fnon-call-exceptions
-fno-inline-small-functions -gno-statement-frontiers -c ovslsrlb.c
x86_64-pc-linux-gnu-gcc-12.0.1: error: ovslsrlb.c: '-fcompare-debug' failure
(length)

For x86_64 the difference looks like this, though the failure is not
target-specific:

--- ovslsrlb.c.gkd      2022-01-28 09:42:43.969435346 +0700
+++ ovslsrlb.gk.c.gkd   2022-01-28 09:42:43.987435394 +0700
@@ -40,8 +40,8 @@
 (note # 0 0 [bb 2] NOTE_INSN_BASIC_BLOCK)
 (note # 0 0 NOTE_INSN_PROLOGUE_END)
 (note # 0 0 NOTE_INSN_FUNCTION_BEG)
-(insn:TI # 0 0 2 (set (mem/c:SI (symbol_ref:DI ("n") [flags 0x2]  <var_decl #
n>) [ MEM[(int *)&n]+0 S4 A32])
-        (const_int 1 [0x1])) "ovslsrlb.c":8:5# {*movsi_internal}
+(insn:TI # 0 0 2 (set (mem/c:SI (symbol_ref:DI ("n") [flags 0x2]  <var_decl #
n>) [ n+0 S4 A32])
+        (const_int 1 [0x1])) "ovslsrlb.c":17:5# {*movsi_internal}
      (nil))
 (insn # 0 0 2 (parallel [
             (set (reg:DI 0 ax)

[Bug debug/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |10.4
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2022-01-28
           Priority|P3                          |P2
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  Caused by store-merging:

-Coalescing successful!
-Merged into 1 stores
-New sequence of 1 stores to replace old one of 2 stores
-Merging successful!
 __attribute__((returns_twice, pure))
 int foo ()
 {
   <bb 2> [local count: 1073741824]:
+  # DEBUG BEGIN_STMT
+  # DEBUG INLINE_ENTRY bar
+  # DEBUG BEGIN_STMT
+  # DEBUG BEGIN_STMT
+  n = 0;
+  # DEBUG BEGIN_STMT
+  # DEBUG a => 0
+  # DEBUG BEGIN_STMT

   <bb 3> [local count: 1073312329]:
-  MEM[(int *)&n] = 1;
+  n = 1;
+  # DEBUG BEGIN_STMT

   <bb 4> [local count: 1073741824]:
+  # DEBUG BEGIN_STMT
   return 0;

   <bb 5> [local count: 429496]:

so we fail to merge n = 0 and n = 1 with -g.

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Note we have an abnormal edge from BB2 to the abnormal dispatcher (missed
optimization, we should get rid of it), but the last stmts in the abnormal
edge sources are debug vs. non-debug stmts (but neither are calls which
are the only sources of abnormal control flow - thus the missed optimization,
to be rectified in CFG cleanup I guess or some pass missing to call
abnormal cleanup - that would be IPA inlining in this case).

int foo ()
{
;;   basic block 2, loop depth 0
;;    pred:       ENTRY
  # DEBUG BEGIN_STMT
  # DEBUG INLINE_ENTRY bar
  # DEBUG BEGIN_STMT
  # DEBUG BEGIN_STMT
  n = 0;
  # DEBUG BEGIN_STMT
  # DEBUG a => 0
  # DEBUG BEGIN_STMT
;;    succ:       3
;;                5

;;   basic block 3, loop depth 0
;;    pred:       2
  n = 1;
  # DEBUG BEGIN_STMT
;;    succ:       4

;;   basic block 4, loop depth 0
;;    pred:       3
;;                5
  # DEBUG BEGIN_STMT
  return 0;
;;    succ:       EXIT

;;   basic block 5, loop depth 0
;;    pred:       2
  .ABNORMAL_DISPATCHER (0);
;;    succ:       4

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I'll have a look.

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[10/11/12 Regression]       |[10/11/12 Regression]
                   |'-fcompare-debug' failure   |'-fcompare-debug' failure
                   |(length) w/ -O2             |(length) w/ -O2
                   |-fnon-call-exceptions       |-fnon-call-exceptions
                   |-fno-inline-small-functions |-fno-inline-small-functions
                   |                            |since
                   |                            |r10-3575-g629387a6586a7531
                 CC|                            |ebotcazou at gcc dot gnu.org,
                   |                            |marxin at gcc dot gnu.org

--- Comment #4 from Martin Liška <marxin at gcc dot gnu.org> ---
Started with r10-3575-g629387a6586a7531.

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
OK, so with the abnormal edges the issue is that DCE1 removes the recursive
call to foo () (because the function is "wrongly" pure) which causes us to
clear
cfun->calls_setjmp but then any further gimple_purge_dead_abnormal_call_edges
will simply do nothing (as premature optimization one could guess).  So the
bug would be that we fail to wipe all abnormal edges when a function no
longer calls setjmp or has a nonlocal label.

The easiest fix is to remove the premature optimization in
gimple_purge_dead_abnormal_call_edges.  That leaves the abnormal edges from
bar() even after DCE removes the recursive call to foo() (still a missed
optimization) but it at least will not leave us with stray abnormal edges
from non-call / non-computed goto stmts in the IL.

It also resolves the compare-debug failure for this testcase (but maybe there's
a latent issue in store-merging still).

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
I'm testing a patch to gimple_purge_dead_abnormal_call_edges, leaving the
store-merging issue to Jakub.

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 52309
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=52309&action=edit
gcc12-pr104263.patch

Untested fix for the store-merging bug.

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[ebotcazou at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #8 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
> Untested fix for the store-merging bug.

FWIW it looks good to me, thanks!

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:5b6f04276e3d1f20817ed37b2e26e43bd12cc0d2

commit r12-6917-g5b6f04276e3d1f20817ed37b2e26e43bd12cc0d2
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Jan 28 10:55:29 2022 +0100

    tree-optimization/104263 - avoid retaining abnormal edges for non-call/goto
stmts

    This removes a premature optimization from
    gimple_purge_dead_abnormal_call_edges which, after eliding the
    last setjmp (or computed goto) statement from a function and
    thus clearing cfun->calls_setjmp, leaves us with the abnormal
    edges from other calls that are elided for example via inlining
    or DCE.  That's a CFG / IL combination that should be impossible
    (not addressing the fact that with cfun->calls_setjmp and
    cfun->has_nonlocal_label cleared we should not have any abnormal
    edge at all).

    For the testcase in the PR this means that IPA inlining will
    remove the abormal edges from the block after inlining the call
    the edge was coming from.

    2022-01-28  Richard Biener  <rguenther@suse.de>

            PR tree-optimization/104263
            * tree-cfg.cc (gimple_purge_dead_abnormal_call_edges):
            Purge edges also when !cfun->has_nonlocal_label
            and !cfun->calls_setjmp.

            * gcc.dg/tree-ssa/inline-13.c: New testcase.

[Bug tree-optimization/104263] [10/11/12 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #10 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:a591c71b41e18e4ff86852a974592af4962aef57

commit r12-6923-ga591c71b41e18e4ff86852a974592af4962aef57
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Fri Jan 28 19:02:26 2022 +0100

    store-merging: Fix up a -fcompare-debug bug in get_status_for_store_merging
[PR104263]

    As mentioned in the PRthe following testcase fails, because the last
    stmt of a bb with -g is a debug stmt and get_status_for_store_merging
    uses gimple_seq_last_stmt (bb_seq (bb)) when testing if it is valid
    for store merging.  The debug stmt isn't valid, while a stmt at that
    position with -g0 is valid and so the divergence.

    As we walk the whole bb already, this patch just remembers the last
    non-debug stmt, so that we don't need to skip backwards debug stmts at the
    end of the bb to find last real stmt.

    2022-01-28  Jakub Jelinek  <jakub@redhat.com>

            PR tree-optimization/104263
            * gimple-ssa-store-merging.cc (get_status_for_store_merging): For
            cfun->can_throw_non_call_exceptions && cfun->eh test whether
            last non-debug stmt in the bb is store_valid_for_store_merging_p
            rather than last stmt.

            * gcc.dg/pr104263.c: New test.

[Bug tree-optimization/104263] [10/11 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[10/11/12 Regression]       |[10/11 Regression]
                   |'-fcompare-debug' failure   |'-fcompare-debug' failure
                   |(length) w/ -O2             |(length) w/ -O2
                   |-fnon-call-exceptions       |-fnon-call-exceptions
                   |-fno-inline-small-functions |-fno-inline-small-functions
                   |since                       |since
                   |r10-3575-g629387a6586a7531  |r10-3575-g629387a6586a7531

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed on the trunk so far.

[Bug tree-optimization/104263] [10/11 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #12 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:73b30d18a04bd6efa7e25c28dac1c863dc1cb06e

commit r11-9595-g73b30d18a04bd6efa7e25c28dac1c863dc1cb06e
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Fri Jan 28 19:02:26 2022 +0100

    store-merging: Fix up a -fcompare-debug bug in get_status_for_store_merging
[PR104263]

    As mentioned in the PRthe following testcase fails, because the last
    stmt of a bb with -g is a debug stmt and get_status_for_store_merging
    uses gimple_seq_last_stmt (bb_seq (bb)) when testing if it is valid
    for store merging.  The debug stmt isn't valid, while a stmt at that
    position with -g0 is valid and so the divergence.

    As we walk the whole bb already, this patch just remembers the last
    non-debug stmt, so that we don't need to skip backwards debug stmts at the
    end of the bb to find last real stmt.

    2022-01-28  Jakub Jelinek  <jakub@redhat.com>

            PR tree-optimization/104263
            * gimple-ssa-store-merging.c (get_status_for_store_merging): For
            cfun->can_throw_non_call_exceptions && cfun->eh test whether
            last non-debug stmt in the bb is store_valid_for_store_merging_p
            rather than last stmt.

            * gcc.dg/pr104263.c: New test.

    (cherry picked from commit a591c71b41e18e4ff86852a974592af4962aef57)

[Bug tree-optimization/104263] [10 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[10/11 Regression]          |[10 Regression]
                   |'-fcompare-debug' failure   |'-fcompare-debug' failure
                   |(length) w/ -O2             |(length) w/ -O2
                   |-fnon-call-exceptions       |-fnon-call-exceptions
                   |-fno-inline-small-functions |-fno-inline-small-functions
                   |since                       |since
                   |r10-3575-g629387a6586a7531  |r10-3575-g629387a6586a7531

--- Comment #13 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed for 11.3+ too.

[Bug tree-optimization/104263] [10 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #14 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Alexandre Oliva <aoliva@gcc.gnu.org>:

https://gcc.gnu.org/g:33c7df5854ecb31e2d645ab672ad3987b591ebfd

commit r12-7383-g33c7df5854ecb31e2d645ab672ad3987b591ebfd
Author: Alexandre Oliva <oliva@adacore.com>
Date:   Thu Feb 24 22:03:38 2022 -0300

    Add testcase from PR103845

    This problem was already fixed as part of PR104263: the abnormal edge
    that remained from before inlining didn't make sense after inlining.
    So this patch adds only the testcase.


    for  gcc/testsuite/ChangeLog

            PR tree-optimization/103845
            PR tree-optimization/104263
            * gcc.dg/pr103845.c: New.

[Bug tree-optimization/104263] [10 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

--- Comment #15 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Jakub Jelinek
<jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:7bb29feb789dfa05297998e1c241d438307cc3d2

commit r10-10671-g7bb29feb789dfa05297998e1c241d438307cc3d2
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Fri Jan 28 19:02:26 2022 +0100

    store-merging: Fix up a -fcompare-debug bug in get_status_for_store_merging
[PR104263]

    As mentioned in the PRthe following testcase fails, because the last
    stmt of a bb with -g is a debug stmt and get_status_for_store_merging
    uses gimple_seq_last_stmt (bb_seq (bb)) when testing if it is valid
    for store merging.  The debug stmt isn't valid, while a stmt at that
    position with -g0 is valid and so the divergence.

    As we walk the whole bb already, this patch just remembers the last
    non-debug stmt, so that we don't need to skip backwards debug stmts at the
    end of the bb to find last real stmt.

    2022-01-28  Jakub Jelinek  <jakub@redhat.com>

            PR tree-optimization/104263
            * gimple-ssa-store-merging.c (get_status_for_store_merging): For
            cfun->can_throw_non_call_exceptions && cfun->eh test whether
            last non-debug stmt in the bb is store_valid_for_store_merging_p
            rather than last stmt.

            * gcc.dg/pr104263.c: New test.

    (cherry picked from commit a591c71b41e18e4ff86852a974592af4962aef57)

[Bug tree-optimization/104263] [10 Regression] '-fcompare-debug' failure (length) w/ -O2 -fnon-call-exceptions -fno-inline-small-functions since r10-3575-g629387a6586a7531

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104263

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

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

--- Comment #16 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed for 10.4 too.