[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

--- Comment #14 from Andrew Pinski  ---
(In reply to Andrew Pinski from comment #13)
> So the testcase gcc.dg/pr109410.c passes but if you add -g, the ICE will
> show up still.
> 
> This also means there will be some compare-debug failure too as the IR/CFG
> will be different with/without -g.
> 
> I suspect gsi_after_labels here should really be
> gsi_start_nondebug_after_labels_bb .

That is:
diff --git a/gcc/tree-ssa-reassoc.cc b/gcc/tree-ssa-reassoc.cc
index 067a3f07f7e..c71780e6d54 100644
--- a/gcc/tree-ssa-reassoc.cc
+++ b/gcc/tree-ssa-reassoc.cc
@@ -1563,7 +1563,7 @@ build_and_add_sum (tree type, tree op1, tree op2, enum
tree_code opcode)
   if ((!op1def || gimple_nop_p (op1def))
   && (!op2def || gimple_nop_p (op2def)))
 {
-  gsi = gsi_after_labels (single_succ (ENTRY_BLOCK_PTR_FOR_FN (cfun)));
+  gsi = gsi_start_nondebug_after_labels_bb (single_succ
(ENTRY_BLOCK_PTR_FOR_FN (cfun)));
   if (!gsi_end_p (gsi)
  && is_gimple_call (gsi_stmt (gsi))
  && (gimple_call_flags (gsi_stmt (gsi)) & ECF_RETURNS_TWICE))

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

Andrew Pinski  changed:

   What|Removed |Added

   Keywords||compare-debug-failure

--- Comment #13 from Andrew Pinski  ---
So the testcase gcc.dg/pr109410.c passes but if you add -g, the ICE will show
up still.

This also means there will be some compare-debug failure too as the IR/CFG will
be different with/without -g.

I suspect gsi_after_labels here should really be
gsi_start_nondebug_after_labels_bb .

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[sjames at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

--- Comment #12 from Sam James  ---
Bisect gives a nonsensical result of r13-7156-g31eb8f18bbe646 with initial good
as r13-7155-g51856718a82ce60f067910d9037ca255645b37eb and bad as
releases/gcc-13.

I've checked out r13-7155-g51856718a82ce60f067910d9037ca255645b37eb and with
that commit, the test case still fails:
```
$ /tmp/bisect/bin/gcc -c -g -O1
/home/sam/git/gcc/gcc/testsuite/gcc.dg/pr109410.c
/home/sam/git/gcc/gcc/testsuite/gcc.dg/pr109410.c: In function ‘foo’:
/home/sam/git/gcc/gcc/testsuite/gcc.dg/pr109410.c:14:1: error: returns_twice
call is not first in basic block 2
   14 | foo (int x, int y)
  | ^~~
baz (x_7(D), y_8(D));
during GIMPLE pass: reassoc
/home/sam/git/gcc/gcc/testsuite/gcc.dg/pr109410.c:14:1: internal compiler
error: verify_flow_info failed
0x9c06ee verify_flow_info()
../.././gcc/cfghooks.cc:285
0xdb8077 execute_function_todo
../.././gcc/passes.cc:2110
0xdb85be execute_todo
../.././gcc/passes.cc:2152
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See  for instructions.
```

with:
```
gcc (GCC) 13.0.1 20230412 (experimental)
Copyright (C) 2023 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
```

so I don't think the original fix is complete?

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[sjames at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

--- Comment #11 from Sam James  ---
Broken for 13 again too:

$ gcc -c -g -O1 ./gcc/testsuite/gcc.dg/pr109410.c
./gcc/testsuite/gcc.dg/pr109410.c: In function ‘foo’:
./gcc/testsuite/gcc.dg/pr109410.c:14:1: error: returns_twice call is not first
in basic block 2
   14 | foo (int x, int y)
  | ^~~
baz (x_7(D), y_8(D));
during GIMPLE pass: reassoc
./gcc/testsuite/gcc.dg/pr109410.c:14:1: internal compiler error:
verify_flow_info failed
0x7a85a2 verify_flow_info()
   
/usr/src/debug/sys-devel/gcc-13.0.1_pre20230421/gcc-13.1.0-RC-20230421/gcc/cfghooks.cc:285
0x19c3a82 execute_function_todo
   
/usr/src/debug/sys-devel/gcc-13.0.1_pre20230421/gcc-13.1.0-RC-20230421/gcc/passes.cc:2110
0x192e2e6 do_per_function
   
/usr/src/debug/sys-devel/gcc-13.0.1_pre20230421/gcc-13.1.0-RC-20230421/gcc/passes.cc:1694
0x192e2e6 execute_todo
   
/usr/src/debug/sys-devel/gcc-13.0.1_pre20230421/gcc-13.1.0-RC-20230421/gcc/passes.cc:2152
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See  for instructions.

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[sjames at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

Sam James  changed:

   What|Removed |Added

 Status|RESOLVED|REOPENED
 Resolution|FIXED   |---

--- Comment #10 from Sam James  ---
Reopening then.

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[dcb314 at hotmail dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

David Binderman  changed:

   What|Removed |Added

 CC||dcb314 at hotmail dot com

--- Comment #9 from David Binderman  ---
I think this one might have returned. With today's gcc trunk:

$ ~/gcc/results/bin/gcc -c  ./gcc.dg/pr109410.c
$ ~/gcc/results/bin/gcc -c -g ./gcc.dg/pr109410.c
$ ~/gcc/results/bin/gcc -c -g -O1 ./gcc.dg/pr109410.c
./gcc.dg/pr109410.c: In function ‘foo’:
./gcc.dg/pr109410.c:14:1: error: returns_twice call is not first in basic block
2
   14 | foo (int x, int y)
  | ^~~
baz (x_7(D), y_8(D));
during GIMPLE pass: reassoc
./gcc.dg/pr109410.c:14:1: internal compiler error: verify_flow_info failed


$ ~/gcc/results/bin/gcc -v
Using built-in specs.
COLLECT_GCC=/home/dcb36/gcc/results/bin/gcc
COLLECT_LTO_WRAPPER=/home/dcb36/gcc/results.20230422/libexec/gcc/x86_64-pc-linux-gnu/14.0.0/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: ../trunk.year/configure
--prefix=/home/dcb36/gcc/results.20230422 --disable-multilib
--disable-bootstrap --with-pkgversion=cda246f8b421ba85 --enable-checking=yes
--enable-languages=c,c++,fortran
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 14.0.0 20230422 (experimental) (cda246f8b421ba85) 
$

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[cvs-commit at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

--- Comment #8 from CVS Commits  ---
The releases/gcc-12 branch has been updated by Jakub Jelinek
:

https://gcc.gnu.org/g:0e55b8e091a7263364cb08fd2c9172babb554ab1

commit r12-9428-g0e55b8e091a7263364cb08fd2c9172babb554ab1
Author: Jakub Jelinek 
Date:   Wed Apr 12 16:55:15 2023 +0200

reassoc: Fix up another ICE with returns_twice call [PR109410]

The following testcase ICEs in reassoc, unlike the last case I've fixed
there here SSA_NAME_USED_IN_ABNORMAL_PHI is not the case anywhere.
build_and_add_sum places new statements after the later appearing
definition
of an operand but if both operands are default defs or constants, we place
statement at the start of the function.

If the very first statement of a function is a call to returns_twice
function, this doesn't work though, because that call has to be the first
thing in its basic block, so the following patch splits the entry successor
edge such that the new statements are added into a different block from the
returns_twice call.

I think we should in stage1 reconsider such placements, I think it
unnecessarily enlarges the lifetime of the new lhs if its operand(s)
are used more than once in the function.  Unless something sinks those
again.  Would be nice to place it closer to the actual uses (or where
they will be placed).

2023-04-12  Jakub Jelinek  

PR tree-optimization/109410
* tree-ssa-reassoc.cc (build_and_add_sum): Split edge from entry
block if first statement of the function is a call to returns_twice
function.

* gcc.dg/pr109410.c: New test.

(cherry picked from commit 51856718a82ce60f067910d9037ca255645b37eb)

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[jakub at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

Jakub Jelinek  changed:

   What|Removed |Added

 Status|NEW |RESOLVED
 Resolution|--- |FIXED

--- Comment #7 from Jakub Jelinek  ---
Fixed (though might be worth backporting).

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[cvs-commit at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

--- Comment #6 from CVS Commits  ---
The master branch has been updated by Jakub Jelinek :

https://gcc.gnu.org/g:51856718a82ce60f067910d9037ca255645b37eb

commit r13-7155-g51856718a82ce60f067910d9037ca255645b37eb
Author: Jakub Jelinek 
Date:   Wed Apr 12 16:55:15 2023 +0200

reassoc: Fix up another ICE with returns_twice call [PR109410]

The following testcase ICEs in reassoc, unlike the last case I've fixed
there here SSA_NAME_USED_IN_ABNORMAL_PHI is not the case anywhere.
build_and_add_sum places new statements after the later appearing
definition
of an operand but if both operands are default defs or constants, we place
statement at the start of the function.

If the very first statement of a function is a call to returns_twice
function, this doesn't work though, because that call has to be the first
thing in its basic block, so the following patch splits the entry successor
edge such that the new statements are added into a different block from the
returns_twice call.

I think we should in stage1 reconsider such placements, I think it
unnecessarily enlarges the lifetime of the new lhs if its operand(s)
are used more than once in the function.  Unless something sinks those
again.  Would be nice to place it closer to the actual uses (or where
they will be placed).

2023-04-12  Jakub Jelinek  

PR tree-optimization/109410
* tree-ssa-reassoc.cc (build_and_add_sum): Split edge from entry
block if first statement of the function is a call to returns_twice
function.

* gcc.dg/pr109410.c: New test.

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[sjames at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

--- Comment #5 from Sam James  ---
(In reply to Jakub Jelinek from comment #4)
> PR108783?

Test case was from it. I don't mind not adding such things to See Also though,
I'm still new to bug wrangling. Sorry if it's wrong!

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[jakub at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

--- Comment #4 from Jakub Jelinek  ---
PR108783?
Anyway, will have a look now.

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[rguenth at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

Richard Biener  changed:

   What|Removed |Added

 CC||jakub at gcc dot gnu.org,
   ||rguenth at gcc dot gnu.org
   Priority|P3  |P2
   Last reconfirmed||2023-04-11
 Ever confirmed|0   |1
 Status|UNCONFIRMED |NEW

--- Comment #3 from Richard Biener  ---
Confirmed.  ISTR we did have another issue like this Jakub fixed recently?

[local count: 1073741824]:
+  _4 = x_7(D) > 41;
   baz (x_7(D), y_8(D));
   goto ; [99.96%]

@@ -101,9 +141,8 @@
[local count: 1073312329]:
   _1 = y_8(D) != 0;
   _2 = x_7(D) == 42;
-  _3 = _1 | _2;
   _12 = x_7(D) > 42;
-  _10 = _3 | _12;
+  _10 = _4 | _1;
   _13 = (int) _10;
   return _13;

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

--- Comment #2 from Andrew Pinski  ---
The broken IR has been there since at least 4.9.0:
  :
  _1 = x_4(D) > 41;
  baz (x_4(D), y_5(D));
  goto ;

  :
  ABNORMAL_DISPATCHER (0);

  :
  _8 = y_5(D) != 0;
  _9 = x_4(D) == 42;
  _11 = x_4(D) > 42;
  _12 = _1 | _8;
  _13 = (int) _12;


4.7.x didn't have the ABNORMAL_DISPATCHER part of the IR either.

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

Andrew Pinski  changed:

   What|Removed |Added

   Keywords||ice-checking

--- Comment #1 from Andrew Pinski  ---
Most likely r13-1754-g7a158a5776f5ca which introduced the checking in the first
place.

[Bug tree-optimization/109410] [13 Regression] ICE: verify_flow_info failed

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109410

Andrew Pinski  changed:

   What|Removed |Added

  Component|c   |tree-optimization
   Target Milestone|--- |13.0