[Bug tree-optimization/109002] New: -O1 -ftree-pre -ftree-partial-pre results in stall value

[Bug tree-optimization/109002] New: -O1 -ftree-pre -ftree-partial-pre results in stall value

[akihiko.odaki at daynix dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109002

            Bug ID: 109002
           Summary: -O1 -ftree-pre -ftree-partial-pre results in stall
                    value
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: akihiko.odaki at daynix dot com
  Target Milestone: ---

The following should emit an executable which prints "a" forever, but the
actual compiled executable prints nothing and just hangs. It seems the code
sees a stall value of variable g.

gcc -O1 -ftree-pre -ftree-partial-pre -x c - <<EOF
#include <stdio.h>

int g;
int h;

int main(void)
{
        for (int i = 0; ; i++) {
                for (int j = 0; j < g; j++);

                if (i & 1) {
                        if (h)
                                continue;

                        if (g)
                                puts("a");

                        g = 1;
                }
        }
}
EOF

I reproduced this bug with the current master. gcc -v says:
Using built-in specs.
COLLECT_GCC=/home/alarm/gcc-installation/usr/local/bin/gcc
COLLECT_LTO_WRAPPER=/home/alarm/gcc-installation/usr/local/bin/../libexec/gcc/aarch64-unknown-linux-gnu/13.0.1/lto-wrapper
Target: aarch64-unknown-linux-gnu
Configured with: ../gcc/configure --disable-bootstrap : (reconfigured)
../gcc/configure --disable-bootstrap
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 13.0.1 20230302 (experimental) (GCC) 
COLLECT_GCC_OPTIONS='-v' '-mlittle-endian' '-mabi=lp64'

/home/alarm/gcc-installation/usr/local/bin/../libexec/gcc/aarch64-unknown-linux-gnu/13.0.1/cc1
-quiet -v -iprefix
/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/
/home/alarm/a.c -quiet -dumpbase a.c -dumpbase-ext .c -mlittle-endian
-mabi=lp64 -version -o /tmp/ccetU9yt.s
GNU C17 (GCC) version 13.0.1 20230302 (experimental)
(aarch64-unknown-linux-gnu)
        compiled by GNU C version 12.1.0, GMP version 6.2.1, MPFR version
4.2.0, MPC version 1.3.1, isl version isl-0.25-GMP

GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096
ignoring nonexistent directory
"/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/../../../../aarch64-unknown-linux-gnu/include"
ignoring duplicate directory
"/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/../../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/include"
ignoring duplicate directory
"/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/../../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/include-fixed"
ignoring nonexistent directory
"/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/../../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/../../../../aarch64-unknown-linux-gnu/include"
#include "..." search starts here:
#include <...> search starts here:

/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/include

/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/include-fixed
 /usr/local/include
 /usr/include
End of search list.
Compiler executable checksum: 374b2759695ed1b72f52cb0f1847fbb3
COLLECT_GCC_OPTIONS='-v' '-mlittle-endian' '-mabi=lp64'
 as -v -EL -mabi=lp64 -o /tmp/ccTu98hj.o /tmp/ccetU9yt.s
GNU assembler version 2.38 (aarch64-unknown-linux-gnu) using BFD version (GNU
Binutils) 2.38
COMPILER_PATH=/home/alarm/gcc-installation/usr/local/bin/../libexec/gcc/aarch64-unknown-linux-gnu/13.0.1/:/home/alarm/gcc-installation/usr/local/bin/../libexec/gcc/
LIBRARY_PATH=/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/:/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/:/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/../../../../lib64/:/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/../../../:/lib/:/usr/lib/
COLLECT_GCC_OPTIONS='-v' '-mlittle-endian' '-mabi=lp64' '-dumpdir' 'a.'

/home/alarm/gcc-installation/usr/local/bin/../libexec/gcc/aarch64-unknown-linux-gnu/13.0.1/collect2
-plugin
/home/alarm/gcc-installation/usr/local/bin/../libexec/gcc/aarch64-unknown-linux-gnu/13.0.1/liblto_plugin.so
-plugin-opt=/home/alarm/gcc-installation/usr/local/bin/../libexec/gcc/aarch64-unknown-linux-gnu/13.0.1/lto-wrapper
-plugin-opt=-fresolution=/tmp/ccup8iGA.res -plugin-opt=-pass-through=-lgcc
-plugin-opt=-pass-through=-lgcc_s -plugin-opt=-pass-through=-lc
-plugin-opt=-pass-through=-lgcc -plugin-opt=-pass-through=-lgcc_s
--eh-frame-hdr -dynamic-linker /lib/ld-linux-aarch64.so.1 -X -EL -maarch64linux
/lib/crt1.o /lib/crti.o
/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/crtbegin.o
-L/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1
-L/home/alarm/gcc-installation/usr/local/bin/../lib/gcc
-L/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/../../../../lib64
-L/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/../../..
/tmp/ccTu98hj.o -lgcc --push-state --as-needed -lgcc_s --pop-state -lc -lgcc
--push-state --as-needed -lgcc_s --pop-state
/home/alarm/gcc-installation/usr/local/bin/../lib/gcc/aarch64-unknown-linux-gnu/13.0.1/crtend.o
/lib/crtn.o
COLLECT_GCC_OPTIONS='-v' '-mlittle-endian' '-mabi=lp64' '-dumpdir' 'a.'

git bisect revealed this is a regression caused by commit
5edf02ed2b6de024f83a023d046a6a18f645bc83.

[Bug tree-optimization/109002] -O1 -ftree-pre -ftree-partial-pre results in stall value

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109002

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note there will be undefined behavior when i become INT_MAx.

[Bug tree-optimization/109002] -O1 -ftree-pre -ftree-partial-pre results in stall value

[akihiko.odaki at daynix dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109002

--- Comment #2 from Akihiko Odaki <akihiko.odaki at daynix dot com> ---
Oops. Replacing i++ with i = !i removes the undefined behavior while the bug
still remains.

[Bug tree-optimization/109002] -O1 -ftree-pre -ftree-partial-pre results in stall value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109002

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2023-03-03
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
I will have a look.

[Bug tree-optimization/109002] -O1 -ftree-pre -ftree-partial-pre results in stall value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109002

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
When doing partial PRE we somehow lose the effect of

  g = 1;

we also generate weird PHIs:

<bb 2>
  pretmp_20 = h;
  pretmp_22 = g; 


<bb 3>
  # prephitmp_21 = PHI <pretmp_20(2), prephitmp_25(10)>
  # prephitmp_23 = PHI <pretmp_22(2), prephitmp_24(10)>
  # prephitmp_24 = PHI <pretmp_22(2), prephitmp_24(10)>
  # prephitmp_25 = PHI <pretmp_20(2), prephitmp_25(10)>

those are all unnecessary ...

There's one old detail I never understood in partial ANTIC compute and that
gets us weird expressions in here, which is that we do not PHI translate
ANTIC_IN when computing PA_OUT.  Fixing that fixes this bug.

I have no idea how to create a testcase for the testsuite though besides
verifying we're not inserting any PHI - but we also fail to partial-PRE the
load
of h here, likely because of the iteration order imposed by the infinite
loop which is probably also key to the miscompilation.

Anyway, I have a fix, but I'm not 100% sure it fixes the actual cause.

[Bug tree-optimization/109002] [13 Regression] -O1 -ftree-pre -ftree-partial-pre results in stall value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109002

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|-O1 -ftree-pre              |[13 Regression] -O1
                   |-ftree-partial-pre results  |-ftree-pre
                   |in stall value              |-ftree-partial-pre results
                   |                            |in stall value
   Target Milestone|---                         |13.0
      Known to work|                            |12.2.0
           Priority|P3                          |P1
           Keywords|                            |needs-bisection

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Looks like GCC 12 "works", I suspect the compile-time optimizations in pruning
to be the reason since PA_IN has bogus virtual operands in expressions which is
what the fix fixes.

[Bug tree-optimization/109002] [13 Regression] -O1 -ftree-pre -ftree-partial-pre results in stall value

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109002

--- Comment #6 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:0132acc03cada2c3b47c48a205e821563153fc80

commit r13-6443-g0132acc03cada2c3b47c48a205e821563153fc80
Author: Richard Biener <rguenther@suse.de>
Date:   Fri Mar 3 10:41:29 2023 +0100

    tree-optimization/109002 - partial PRE miscompilation

    Partial PRE ends up miscompiling the testcase in PR109002, likely
    involving a corner case when inifinite loops are involved.  The
    following avoids the miscompilation by addressing a long-standing
    oddity that manifests in odd partial partial redundancies eliminated
    that are full redundancies.  The oddity is that while we properly
    PHI translate the PA_IN set from the successors when computing
    PA_OUT but we fail to do the same for ANTIC_IN which is supposed
    to be unioned.  That results in expressions with wrong virtual
    operands being placed in the PA_OUT/IN sets and the pruning
    machinery to go wrong because it assumes the expressions in the
    sets have virtual operands that are valid in the respective blocks.

            PR tree-optimization/109002
            * tree-ssa-pre.cc (compute_partial_antic_aux): Properly
            PHI-translate ANTIC_IN.

            * gcc.dg/torture/pr109002.c: New testcase.

[Bug tree-optimization/109002] [13 Regression] -O1 -ftree-pre -ftree-partial-pre results in stall value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109002

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

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

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.