[Bug c/113396] New: csmith: differences from -O2 to -O3

[Bug c/113396] New: csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

            Bug ID: 113396
           Summary: csmith: differences from -O2 to -O3
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dcb314 at hotmail dot com
  Target Milestone: ---

Created attachment 57082
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57082&action=edit
C source code

The attached Csmith generated code does this:

foundBugs $ ~/gcc/results/bin/gcc -w bug998.c
foundBugs $ valgrind -q ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results/bin/gcc -w -O2 bug998.c
foundBugs $ valgrind -q ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results/bin/gcc -w -O3 bug998.c
foundBugs $ valgrind -q ./a.out
checksum = 130B5204
foundBugs $ ~/gcc/results/bin/gcc -w -O3 -fno-strict-aliasing
-fno-loop-interchange bug998.c
foundBugs $ valgrind -q ./a.out
checksum = 130B5204
foundBugs $

[Bug c/113396] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #1 from David Binderman <dcb314 at hotmail dot com> ---
foundBugs $ ~/gcc/results/bin/gcc -w  -O2  bug998.c -o two.exe
foundBugs $ ~/gcc/results/bin/gcc -w  -O3  bug998.c -o three.exe
foundBugs $ ./two.exe 1 > two.out
foundBugs $ ./three.exe 1 > three.out
foundBugs $ diff two.out three.out  | head -20
510,520c510,520
< ...checksum after hashing g_179.f2 : F8C8E936
< ...checksum after hashing g_179.f3 : 22C7AEC4
< ...checksum after hashing g_179.f4 : E057A6AE
< ...checksum after hashing g_179.f5 : 494E34F0
< ...checksum after hashing g_179.f6 : 72F55A7C
< ...checksum after hashing g_179.f7 : 9B53D63F
< ...checksum after hashing g_179.f8 : DD10C49F
< ...checksum after hashing g_179.f9 : 979E03A7
< ...checksum after hashing g_195 : 63EDBCE7
< ...checksum after hashing g_196 : C1F2B26B
< ...checksum after hashing g_226[i][j] : 835FC1
---
> ...checksum after hashing g_179.f2 : A0B13872
> ...checksum after hashing g_179.f3 : A6EAB221
> ...checksum after hashing g_179.f4 : 744D7BAB
> ...checksum after hashing g_179.f5 : 4D71F83C
> ...checksum after hashing g_179.f6 : 54EDEE43
> ...checksum after hashing g_179.f7 : B280A60F
> ...checksum after hashing g_179.f8 : ECCE643D
foundBugs $

[Bug c/113396] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #2 from David Binderman <dcb314 at hotmail dot com> ---
The bug seems to exist from sometime before
g:d042f118798ae0648b45f97e63b0e5ab7c82c8ef, dated 20230205.

[Bug c/113396] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #3 from David Binderman <dcb314 at hotmail dot com> ---
Created attachment 57089
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57089&action=edit
C source code

Partly reduced C source code.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |13.3

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #4 from David Binderman <dcb314 at hotmail dot com> ---
foundBugs $ ../results.20220116/bin/gcc -w -O2 bug998.c
foundBugs $ ./a.out
checksum = 77A231E6
foundBugs $ ../results.20220116/bin/gcc -w -O3 bug998.c
foundBugs $ ./a.out
checksum = 77A231E6
foundBugs $ ../results.20230205/bin/gcc -w -O2 bug998.c
foundBugs $ ./a.out
checksum = 77A231E6
foundBugs $ ../results.20230205/bin/gcc -w -O3 bug998.c
foundBugs $ ./a.out
checksum = 130B5204
foundBugs $ 

So it goes wrong sometime between g:90045c5df5b3c8853e7740fb72a11aead1c489bb
and g:d042f118798ae0648b45f97e63b0e5ab7c82c8ef, a distance of 7403 commits.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #5 from David Binderman <dcb314 at hotmail dot com> ---
Bisect range seems to be g:e03a0a4d73a478928b26213363fa5dbb9fc8695f ..
g:4e1914625dec4aa09a5671c6294e877dbf4518f5, which is 1850 commits.

I will continue the bisection.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #6 from David Binderman <dcb314 at hotmail dot com> ---
Reduced bisect range seems to be g:2c11662391bafd74c9d19bf7626b7bcef41c1323 ..
g:9e0d5db3e04afd2d030ace4ccb5c1af5e9f05a8f, which is 462 commits.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #7 from David Binderman <dcb314 at hotmail dot com> ---
Current range seems to be g:54a5f478487a955c3ffaec3e9164a72599bc1cfb ..
g:1edfc8f2d3307a3ffa077a605f432832d7715462, which is 4 commits.

Of those 4, this one

commit 8c99e307b20c502e55c425897fb3884ba8f05882
Author: Aldy Hernandez <aldyh@redhat.com>
Date:   Sat Jun 25 18:58:02 2022 -0400

looks a likely candidate.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

David Binderman <dcb314 at hotmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |aldyh at redhat dot com

--- Comment #8 from David Binderman <dcb314 at hotmail dot com> ---
As expected:

$ git bisect good fc259b522c0f8b7bbca8e7adcd3da63330094a34
8c99e307b20c502e55c425897fb3884ba8f05882 is the first bad commit
commit 8c99e307b20c502e55c425897fb3884ba8f05882
Author: Aldy Hernandez <aldyh@redhat.com>
Date:   Sat Jun 25 18:58:02 2022 -0400

Over to Aldy for their best opinion.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

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

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

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
r13-1268-g8c99e307b20c502e55c425897fb3884ba8f05882

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #10 from David Binderman <dcb314 at hotmail dot com> ---
Created attachment 57117
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57117&action=edit
C source code

After many hours, cvise appears incapable of reducing the code 
much beyond this version.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[law at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

Jeffrey A. Law <law at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
                 CC|                            |law at gcc dot gnu.org

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2024-03-15
           Priority|P1                          |P2

--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
13.1 was released and broken, so this can't be P1.  It's also still
unconfirmed. The last reduced testcase does

> ./xgcc -B. t.c
> ./a.out 
...checksum after hashing  : FFFFFFFF6522DF69
Segmentation fault (core dumped)
Program received signal SIGSEGV, Segmentation fault.
0x0000000000400792 in main () at t.c:115
115           transparent_crc(g_7[main_i][main_j], "");
(gdb) p main_i
$1 = -1747789432

so it seems a useless reduction.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #12 from Richard Biener <rguenth at gcc dot gnu.org> ---
The second testcase behaves the same with -O0, -O2 and -O3 on x86_64-linux for
me (and with trunk and GCC 13.2.1)

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #13 from David Binderman <dcb314 at hotmail dot com> ---
I had another look at the original source code and got this with
recent gcc trunk:

foundBugs $ ~/gcc/results/bin/gcc -w bug998.c && ./a.out
checksum = 77A231E6

foundBugs $ ~/gcc/results/bin/gcc -w -O2 bug998.c && ./a.out
checksum = 77A231E6

foundBugs $ ~/gcc/results/bin/gcc -w -O3 bug998.c && ./a.out
checksum = 130B5204

foundBugs $ ~/gcc/results/bin/gcc -w -O3 -fno-strict-aliasing
-fno-loop-interchange bug998.c && ./a.out
checksum = 130B5204

foundBugs $ ~/gcc/results/bin/gcc -w -O3 -fno-strict-aliasing
-fno-loop-interchange bug998.c && valgrind -q ./a.out
checksum = 130B5204

foundBugs $ ~/gcc/results/bin/gcc -w -O3 -fno-strict-aliasing
-fno-loop-interchange -fsanitize=address bug998.c &&  ./a.out
checksum = 77A231E6

foundBugs $ ~/gcc/results/bin/gcc -w -O3 -fno-strict-aliasing
-fno-loop-interchange -fsanitize=undefined bug998.c &&  ./a.out
checksum = 77A231E6
foundBugs $ 

Case 3 looks suspicious. I guess if I knew the which flags -O3 switches on,
that aren't in -O2, I could find out which ones are causing trouble.
I don't know a -O2.5

Case 4 makes it look like aliasing and loop interchange aren't at fault.

Cases 5, 6 & 7 show that the usual tools find nothing wrong.

Cases 6 & 7 make it looks like switching on a sanitizer makes the code work. 
I don't understand why that would be.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #14 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
    /* -O3 optimizations.  */
    { OPT_LEVELS_3_PLUS, OPT_fgcse_after_reload, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_fipa_cp_clone, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_floop_interchange, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_floop_unroll_and_jam, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_fpeel_loops, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_fpredictive_commoning, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_fsplit_loops, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_fsplit_paths, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_ftree_loop_distribution, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_ftree_partial_pre, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_funswitch_loops, NULL, 1 },
    { OPT_LEVELS_3_PLUS, OPT_fvect_cost_model_, NULL, VECT_COST_MODEL_DYNAMIC
},
    { OPT_LEVELS_3_PLUS, OPT_fversion_loops_for_strides, NULL, 1 },

    /* -O3 parameters.  */
    { OPT_LEVELS_3_PLUS, OPT__param_max_inline_insns_auto_, NULL, 30 },
    { OPT_LEVELS_3_PLUS, OPT__param_early_inlining_insns_, NULL, 14 },
    { OPT_LEVELS_3_PLUS, OPT__param_inline_heuristics_hint_percent_, NULL, 600
},
    { OPT_LEVELS_3_PLUS, OPT__param_inline_min_speedup_, NULL, 15 },
    { OPT_LEVELS_3_PLUS, OPT__param_max_inline_insns_single_, NULL, 200 },

So, that is -O2 -fgcse-after-reload -fipa-cp-clone -floop-interchange
-floop-unroll-and-jam -fpeel-loops -fpredictive-commoning -fsplit-loops
-fsplit-paths -ftree-loop-distribution -ftree-partial-pre -funswitch-loops
-fvect-cost-model=dynamic -fversion-loops-for-strides
--param=max-inline-insns-auto=30 --param=early-inlining-insns=14
--param=inline-heuristics-hint-percent=600 --param=inline-min-speedup=15
--param=max-inline-insns-single=200
Of course, some further tweaks could be hidden in target files or something
could just test optimize >= 3 or similar.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #15 from David Binderman <dcb314 at hotmail dot com> ---
(In reply to Jakub Jelinek from comment #14)
> So, that is -O2 -fgcse-after-reload -fipa-cp-clone -floop-interchange
> -floop-unroll-and-jam -fpeel-loops -fpredictive-commoning -fsplit-loops
> -fsplit-paths -ftree-loop-distribution -ftree-partial-pre -funswitch-loops
> -fvect-cost-model=dynamic -fversion-loops-for-strides
> --param=max-inline-insns-auto=30 --param=early-inlining-insns=14
> --param=inline-heuristics-hint-percent=600 --param=inline-min-speedup=15
> --param=max-inline-insns-single=200

Thanks for that. None of the -f flags seems to affect anything.

foundBugs $ cat flag.list
--param=early-inlining-insns=14
--param=inline-heuristics-hint-percent=600
--param=inline-min-speedup=15
--param=max-inline-insns-auto=30
--param=max-inline-insns-single=200
-O2
foundBugs $ ~/gcc/results/bin/gcc -w -O2 bug998.c &&  ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results/bin/gcc -w `cat flag.list` bug998.c &&  ./a.out
checksum = 130B5204
foundBugs $ 

So it looks like one or more of the --param flags is to blame.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #16 from David Binderman <dcb314 at hotmail dot com> ---
(In reply to David Binderman from comment #15)
> So it looks like one or more of the --param flags is to blame.

foundBugs $ ~/gcc/results/bin/gcc -w -O2 bug998.c &&  ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results/bin/gcc -w -O2 --param=max-inline-insns-auto=30
bug998.c &&  ./a.out
checksum = 130B5204
foundBugs $ 

It seems to break sometime between 23 and 24.

foundBugs $ ~/gcc/results/bin/gcc -w -O2 --param=max-inline-insns-auto=23
bug998.c &&  ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results/bin/gcc -w -O2 --param=max-inline-insns-auto=24
bug998.c &&  ./a.out
checksum = 130B5204
foundBugs $

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #17 from David Binderman <dcb314 at hotmail dot com> ---
I tried out gcc-13.2 and got the following results:

foundBugs $ ~/gcc/results.13.2.asan.ubsan/bin/gcc -w -O2
--param=max-inline-insns-auto=23 bug998.c &&  valgrind -q ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results.13.2.asan.ubsan/bin/gcc -w -O2
--param=max-inline-insns-auto=24 bug998.c &&  valgrind -q ./a.out
checksum = 130B5204
foundBugs $ 

so it is something that has been going wrong since before gcc-13.2.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #18 from David Binderman <dcb314 at hotmail dot com> ---
(In reply to David Binderman from comment #17)
> I tried out gcc-13.2 and got the following results:
> 
> foundBugs $ ~/gcc/results.13.2.asan.ubsan/bin/gcc -w -O2
> --param=max-inline-insns-auto=23 bug998.c &&  valgrind -q ./a.out
> checksum = 77A231E6
> foundBugs $ ~/gcc/results.13.2.asan.ubsan/bin/gcc -w -O2
> --param=max-inline-insns-auto=24 bug998.c &&  valgrind -q ./a.out
> checksum = 130B5204
> foundBugs $ 
> 
> so it is something that has been going wrong since before gcc-13.2.

Similarly with 13.1:

foundBugs $ ~/gcc/results.13.1.asan.ubsan/bin/gcc -w -O2
--param=max-inline-insns-auto=23 bug998.c &&  valgrind -q ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results.13.1.asan.ubsan/bin/gcc -w -O2
--param=max-inline-insns-auto=24 bug998.c &&  valgrind -q ./a.out
checksum = 130B5204
foundBugs $

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #19 from David Binderman <dcb314 at hotmail dot com> ---
gcc 12.3 seems to get it right:

foundBugs $ ~/gcc/results.12.3.asan.ubsan/bin/gcc -w -O2
--param=max-inline-insns-auto=23 bug998.c &&  valgrind -q ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results.12.3.asan.ubsan/bin/gcc -w -O2
--param=max-inline-insns-auto=24 bug998.c &&  valgrind -q ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results.12.3.asan.ubsan/bin/gcc -w -O2
--param=max-inline-insns-auto=30 bug998.c &&  valgrind -q ./a.out
checksum = 77A231E6
foundBugs $ ~/gcc/results.12.3.asan.ubsan/bin/gcc -w -O3  bug998.c &&  valgrind
-q ./a.out
checksum = 77A231E6
foundBugs $

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

Sam James <sjames at gcc dot gnu.org> changed:

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

--- Comment #20 from Sam James <sjames at gcc dot gnu.org> ---
I'm going to throw my usual cvise setup for miscompiles at it for a few days
and see if anything useful falls out.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #21 from Sam James <sjames at gcc dot gnu.org> ---
Created attachment 57727
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57727&action=edit
reduced.c

This aborts for me at >= -O1 but I'm not convinced it's right.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #22 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Adding  -fno-tree-loop-im allows it to work.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #23 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 57728
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57728&action=edit
Reduced slightly more, removes the string compare

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #57727|0                           |1
        is obsolete|                            |
  Attachment #57728|0                           |1
        is obsolete|                            |

--- Comment #24 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 57729
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57729&action=edit
More reduced

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW

--- Comment #25 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #57729|0                           |1
        is obsolete|                            |

--- Comment #26 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 57730
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57730&action=edit
Slightly cleaned up

Note p has to be __int128, otherwise it works.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|needs-reduction             |needs-bisection

--- Comment #27 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
It might be useful to redo the bisect on the testcase in comment #26 at -O1 .

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #28 from Sam James <sjames at gcc dot gnu.org> ---
Running bisect

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3 since r13-1268-g8c99e307b20c50

[sjames at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

Sam James <sjames at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amacleod at redhat dot com
            Summary|[13/14 Regression] csmith:  |[13/14 Regression] csmith:
                   |differences from -O2 to -O3 |differences from -O2 to -O3
                   |                            |since
                   |                            |r13-1268-g8c99e307b20c50
           Keywords|needs-bisection             |

--- Comment #29 from Sam James <sjames at gcc dot gnu.org> ---
r13-1268-g8c99e307b20c50 again

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3 since r13-1268-g8c99e307b20c50

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

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

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

--- Comment #30 from Richard Biener <rguenth at gcc dot gnu.org> ---
It's LIM2 which applies store-motion to m[0] but keeps m[p] in the loop.
Then with range-info for the int128 of [0, +INF] we get into
get_ref_base_and_extent:

                query = get_range_query (cfun);

                if (TREE_CODE (index) == SSA_NAME
                    && (low_bound = array_ref_low_bound (exp),
                        poly_int_tree_p (low_bound))
                    && (unit_size = array_ref_element_size (exp),
                        TREE_CODE (unit_size) == INTEGER_CST)
                    && query->range_of_expr (vr, index)
                    && !vr.varying_p ()
                    && !vr.undefined_p ())
                  {
                    wide_int min = vr.lower_bound ();
                    wide_int max = vr.upper_bound ();
                    poly_offset_int lbound = wi::to_poly_offset (low_bound);
                    /* Try to constrain maxsize with range information.  */
                    offset_int omax
                      = offset_int::from (max, TYPE_SIGN (TREE_TYPE (index)));
                    if (known_lt (lbound, omax))
                      {
                        poly_offset_int rmaxsize;
                        rmaxsize = (omax - lbound + 1)
                            * wi::to_offset (unit_size) << LOG2_BITS_PER_UNIT;
                        if (!known_size_p (maxsize)
                            || known_lt (rmaxsize, maxsize))

and compute rmaxsize == 0.

What's definitely questionable here is the offset_int::from, though it
fits.  But omax - lbound + 1 overflows to zero here.  It's probably
better to use widest-int here, even if that's possibly slower than
offset_int.  Or ensure offset_int has one more bit than required for
the index type (offset_int has 64bit + 3bits, so 128bits).

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3 since r13-1268-g8c99e307b20c50

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #31 from Richard Biener <rguenth at gcc dot gnu.org> ---
diff --git a/gcc/tree-dfa.cc b/gcc/tree-dfa.cc
index cbd3774b21f..1dbd9bd7a00 100644
--- a/gcc/tree-dfa.cc
+++ b/gcc/tree-dfa.cc
@@ -549,7 +549,8 @@ get_ref_base_and_extent (tree exp, poly_int64 *poffset,
                    /* Try to constrain maxsize with range information.  */
                    offset_int omax
                      = offset_int::from (max, TYPE_SIGN (TREE_TYPE (index)));
-                   if (known_lt (lbound, omax))
+                   if (wi::get_precision (max) < ADDR_MAX_PRECISION
+                       && known_lt (lbound, omax))
                      {
                        poly_offset_int rmaxsize;
                        rmaxsize = (omax - lbound + 1)
@@ -567,7 +568,8 @@ get_ref_base_and_extent (tree exp, poly_int64 *poffset,
                    /* Try to adjust bit_offset with range information.  */
                    offset_int omin
                      = offset_int::from (min, TYPE_SIGN (TREE_TYPE (index)));
-                   if (known_le (lbound, omin))
+                   if (wi::get_precision (min) < ADDR_MAX_PRECISION
+                       && known_le (lbound, omin))
                      {
                        poly_offset_int woffset
                          = wi::sext (omin - lbound,


fixes the testcase but I'm not 100% sure the condition is correct.  We
possibly need to cover for the << LOG2_BITS_PER_UNIT done, so maybe
using ADDR_MAX_BITSIZE is better.  I'm going to test a variant with
using <= ADDR_MAX_BITSIZE.

[Bug middle-end/113396] [13/14 Regression] csmith: differences from -O2 to -O3 since r13-1268-g8c99e307b20c50

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #32 from GCC 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:6a55e39bdb1fdb570730c08413ebbe744e493411

commit r14-9565-g6a55e39bdb1fdb570730c08413ebbe744e493411
Author: Richard Biener <rguenther@suse.de>
Date:   Tue Mar 19 15:25:16 2024 +0100

    middle-end/113396 - int128 array index and value-ranges

    The following fixes bogus truncation of a value-range for an int128
    array index when computing the maximum extent for a variable array
    reference.  Instead of possibly slowing things down by using
    widest_int the following makes sure the range bounds fit within
    the constraints offset_int were designed for.

            PR middle-end/113396
            * tree-dfa.cc (get_ref_base_and_extent): Use index range
            bounds only if they fit within the address-range constraints
            of offset_int.

            * gcc.dg/torture/pr113396.c: New testcase.

[Bug middle-end/113396] [13 Regression] csmith: differences from -O2 to -O3 since r13-1268-g8c99e307b20c50

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |14.0
            Summary|[13/14 Regression] csmith:  |[13 Regression] csmith:
                   |differences from -O2 to -O3 |differences from -O2 to -O3
                   |since                       |since
                   |r13-1268-g8c99e307b20c50    |r13-1268-g8c99e307b20c50

--- Comment #33 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed on trunk sofar.

[Bug middle-end/113396] [13 Regression] csmith: differences from -O2 to -O3 since r13-1268-g8c99e307b20c50

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

--- Comment #34 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-13 branch has been updated by Richard Biener
<rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:7baefcb0a358a47a7e2340432d49f29db798a200

commit r13-8718-g7baefcb0a358a47a7e2340432d49f29db798a200
Author: Richard Biener <rguenther@suse.de>
Date:   Tue Mar 19 15:25:16 2024 +0100

    middle-end/113396 - int128 array index and value-ranges

    The following fixes bogus truncation of a value-range for an int128
    array index when computing the maximum extent for a variable array
    reference.  Instead of possibly slowing things down by using
    widest_int the following makes sure the range bounds fit within
    the constraints offset_int were designed for.

            PR middle-end/113396
            * tree-dfa.cc (get_ref_base_and_extent): Use index range
            bounds only if they fit within the address-range constraints
            of offset_int.

            * gcc.dg/torture/pr113396.c: New testcase.

    (cherry picked from commit 6a55e39bdb1fdb570730c08413ebbe744e493411)

[Bug middle-end/113396] [13 Regression] csmith: differences from -O2 to -O3 since r13-1268-g8c99e307b20c50

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113396

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |13.2.0
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED
      Known to work|                            |13.2.1

--- Comment #35 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.