misc: warnings on MPI_STATUS_IGNORE etc with gcc-11.2.1 #5687

balay · 2021-11-19T17:26:39Z

I'm not sure if this is a compiler issue - or code isse [occurs with mpich - but not openmpi]

This is a 32bit build with gcc (GCC) 11.2.1 20210728 (Red Hat 11.2.1-1) on Fedora 35

[glci@pj03 petsc]$ make V=1 PETSC_ARCH=arch-ci-linux-gcc-complex-opt-32bit arch-ci-linux-gcc-complex-opt-32bit/obj/sys/utils/mpits.o
/home/glci/petsc-hash-pkgs/494b06380/bin/mpicc -c -m32 -fPIC -Wall -Wwrite-strings -Wno-strict-aliasing -Wno-unknown-pragmas -g -O    -I/home/glci/petsc/include -I/home/glci/petsc/arch-ci-linux-gcc-complex-opt-32bit/include -I/home/glci/petsc-hash-pkgs/494b06380/include    -MMD -MP /home/glci/petsc/src/sys/utils/mpits.c -o arch-ci-linux-gcc-complex-opt-32bit/obj/sys/utils/mpits.o
/home/glci/petsc/src/sys/utils/mpits.c: In function ‘PetscCommBuildTwoSidedFReq_Ibarrier’:
/home/glci/petsc/src/sys/utils/mpits.c:425:14: warning: ‘MPI_Testall’ accessing 20 bytes in a region of size 0 [-Wstringop-overflow=]
  425 |       ierr = MPI_Testall(nsends,sendreqs,&sent,MPI_STATUSES_IGNORE);CHKERRMPI(ierr);
      |              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/glci/petsc/src/sys/utils/mpits.c:425:14: note: referencing argument 4 of type ‘MPI_Status *’
In file included from /home/glci/petsc/include/petscsys.h:162,
                 from /home/glci/petsc/src/sys/utils/mpits.c:1:
/home/glci/petsc-hash-pkgs/494b06380/include/mpi.h:947:5: note: in a call to function ‘MPI_Testall’
  947 | int MPI_Testall(int count, MPI_Request array_of_requests[], int *flag,
      |     ^~~~~~~~~~~
/home/glci/petsc/src/sys/utils/mpits.c: In function ‘PetscCommBuildTwoSided_Ibarrier’:
/home/glci/petsc/src/sys/utils/mpits.c:125:14: warning: ‘MPI_Testall’ accessing 20 bytes in a region of size 0 [-Wstringop-overflow=]
  125 |       ierr = MPI_Testall(nsends,sendreqs,&sent,MPI_STATUSES_IGNORE);CHKERRMPI(ierr);
      |              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/glci/petsc/src/sys/utils/mpits.c:125:14: note: referencing argument 4 of type ‘MPI_Status *’
In file included from /home/glci/petsc/include/petscsys.h:162,
                 from /home/glci/petsc/src/sys/utils/mpits.c:1:
/home/glci/petsc-hash-pkgs/494b06380/include/mpi.h:947:5: note: in a call to function ‘MPI_Testall’
  947 | int MPI_Testall(int count, MPI_Request array_of_requests[], int *flag,
      |     ^~~~~~~~~~~
In file included from /home/glci/petsc/include/petscsys.h:1674,
                 from /home/glci/petsc/src/sys/utils/mpits.c:1:
/home/glci/petsc/src/sys/utils/mpits.c: In function ‘PetscCommBuildTwoSidedF’:
/home/glci/petsc/include/petsclog.h:551:81: warning: ‘MPI_Waitall’ accessing 20 bytes in a region of size 0 [-Wstringop-overflow=]
  551 |   ((petsc_wait_all_ct++,petsc_sum_of_waits_ct += (PetscLogDouble) (count),0) || MPI_Waitall((count),(array_of_requests),(array_of_statuses)))
      |                                                                                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/glci/petsc/src/sys/utils/mpits.c:494:10: note: in expansion of macro ‘MPI_Waitall’
  494 |   ierr = MPI_Waitall(nto*ntags,toreqs,MPI_STATUSES_IGNORE);CHKERRMPI(ierr);
      |          ^~~~~~~~~~~
/home/glci/petsc/include/petsclog.h:551:81: note: referencing argument 3 of type ‘MPI_Status *’
  551 |   ((petsc_wait_all_ct++,petsc_sum_of_waits_ct += (PetscLogDouble) (count),0) || MPI_Waitall((count),(array_of_requests),(array_of_statuses)))
      |                                                                                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/glci/petsc/src/sys/utils/mpits.c:494:10: note: in expansion of macro ‘MPI_Waitall’
  494 |   ierr = MPI_Waitall(nto*ntags,toreqs,MPI_STATUSES_IGNORE);CHKERRMPI(ierr);
      |          ^~~~~~~~~~~
In file included from /home/glci/petsc/include/petscsys.h:162,
                 from /home/glci/petsc/src/sys/utils/mpits.c:1:
/home/glci/petsc-hash-pkgs/494b06380/include/mpi.h:946:5: note: in a call to function ‘MPI_Waitall’
  946 | int MPI_Waitall(int count, MPI_Request array_of_requests[], MPI_Status array_of_statuses[]) MPICH_API_PUBLIC;
      |     ^~~~~~~~~~~
In file included from /home/glci/petsc/include/petscsys.h:1674,
                 from /home/glci/petsc/src/sys/utils/mpits.c:1:
/home/glci/petsc/include/petsclog.h:551:81: warning: ‘MPI_Waitall’ accessing 20 bytes in a region of size 0 [-Wstringop-overflow=]
  551 |   ((petsc_wait_all_ct++,petsc_sum_of_waits_ct += (PetscLogDouble) (count),0) || MPI_Waitall((count),(array_of_requests),(array_of_statuses)))
      |                                                                                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/glci/petsc/src/sys/utils/mpits.c:495:10: note: in expansion of macro ‘MPI_Waitall’
  495 |   ierr = MPI_Waitall(*nfrom*ntags,fromreqs,MPI_STATUSES_IGNORE);CHKERRMPI(ierr);
      |          ^~~~~~~~~~~
/home/glci/petsc/include/petsclog.h:551:81: note: referencing argument 3 of type ‘MPI_Status *’
  551 |   ((petsc_wait_all_ct++,petsc_sum_of_waits_ct += (PetscLogDouble) (count),0) || MPI_Waitall((count),(array_of_requests),(array_of_statuses)))
      |                                                                                 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/glci/petsc/src/sys/utils/mpits.c:495:10: note: in expansion of macro ‘MPI_Waitall’
  495 |   ierr = MPI_Waitall(*nfrom*ntags,fromreqs,MPI_STATUSES_IGNORE);CHKERRMPI(ierr);
      |          ^~~~~~~~~~~
In file included from /home/glci/petsc/include/petscsys.h:162,
                 from /home/glci/petsc/src/sys/utils/mpits.c:1:
/home/glci/petsc-hash-pkgs/494b06380/include/mpi.h:946:5: note: in a call to function ‘MPI_Waitall’
  946 | int MPI_Waitall(int count, MPI_Request array_of_requests[], MPI_Status array_of_statuses[]) MPICH_API_PUBLIC;
      |     ^~~~~~~~~~~

I see mpi.h has:

#define MPI_STATUSES_IGNORE (MPI_Status *)1

Changing this '0' gets rid of the warning (openmpi uses this) [same issue with 'MPI_STATUS_IGNORE']

#define MPI_STATUSES_IGNORE (MPI_Status *)0

[glci@pj03 petsc]$ make V=1 PETSC_ARCH=arch-ci-linux-gcc-complex-opt-32bit arch-ci-linux-gcc-complex-opt-32bit/obj/sys/utils/mpits.o
/home/glci/petsc-hash-pkgs/494b06380/bin/mpicc -c -m32 -fPIC -Wall -Wwrite-strings -Wno-strict-aliasing -Wno-unknown-pragmas -g -O    -I/home/glci/petsc/include -I/home/glci/petsc/arch-ci-linux-gcc-complex-opt-32bit/include -I/home/glci/petsc-hash-pkgs/494b06380/include    -MMD -MP /home/glci/petsc/src/sys/utils/mpits.c -o arch-ci-linux-gcc-complex-opt-32bit/obj/sys/utils/mpits.o
[glci@pj03 petsc]$ /home/glci/petsc-hash-pkgs/494b06380/bin/mpichversion 
MPICH Version:          3.4.2
MPICH Release date:     Wed May 26 15:51:40 CDT 2021
MPICH Device:           ch3:nemesis
MPICH configure:        --prefix=/home/glci/petsc-hash-pkgs/494b06380 MAKE=/usr/bin/gmake --libdir=/home/glci/petsc-hash-pkgs/494b06380/lib CC=gcc CFLAGS=-m32 -fPIC -g -O AR=/usr/bin/ar ARFLAGS=cr CXX=g++ CXXFLAGS=-m32 -fPIC -g -O -fPIC -std=gnu++17 FFLAGS=-m32 -fPIC -ffree-line-length-0 -g -O -fallow-argument-mismatch FC=gfortran F77=gfortran FCFLAGS=-m32 -fPIC -ffree-line-length-0 -g -O -fallow-argument-mismatch --enable-shared --with-pm=hydra --disable-java --with-device=ch3:nemesis --enable-g=meminit
MPICH CC:       gcc -m32 -fPIC -g -O   -O2
MPICH CXX:      g++ -m32 -fPIC -g -O -fPIC -std=gnu++17  -O2
MPICH F77:      gfortran -m32 -fPIC -ffree-line-length-0 -g -O -fallow-argument-mismatch  -O2
MPICH FC:       gfortran -m32 -fPIC -ffree-line-length-0 -g -O -fallow-argument-mismatch  -O2
MPICH Custom Information: 
[glci@pj03 petsc]$

The text was updated successfully, but these errors were encountered:

balay · 2021-11-20T16:23:11Z

Ah - I see this with regular 64bit builds aswell..

hzhou · 2021-11-22T16:59:15Z

While many compiler warnings are helpful, I hate to be forced to add the complexity of workarounds just to appease the dumb part of the compiler. 😞

I see a few solutions:

Silence the warning with -Wno-stringop-overflow. Not really a solution.
Define the constants to 0. It's only a fix because compiler made this an exception. It cornered us to not able to add additional constants.
Add dummy STATUS objects and points to it. We already do this with MPI_UNWEIGHTED and MPI_WEIGHTS_EMPTY. It is the most semantic correct solution, but the dummy objects add unnecessary (someone refers to it as accidental) complexity.

The solution 3 need #5682 and potentially other patches to work, but it is an un-compromising solution that is most likely future-proof.

balay · 2021-11-22T17:02:03Z

Right now - I'm using -Wno-stringop-overflow [as I'd like to use '-Werror' to catch issues in CI]

hzhou · 2021-11-22T17:05:50Z

@raffenet What do you think regarding #5687 (comment)?

raffenet · 2021-11-22T18:15:21Z

@raffenet What do you think regarding #5687 (comment)?

Unfortunately we're hamstrung by ABI with these values for now. We should definitely add them to the ABI wishlist.

If we do decide to change ABI, I think using 0 or NULL (solution 2) is OK. I guess an advantage to using a dummy struct (solution 3) is that we could throw and error on NULL usage, since that is not allowed by the specification.

raffenet · 2021-11-22T18:21:36Z

Added a bullet onto the ABI wiki: https://github.com/pmodels/mpich/wiki/ABI-Change-Wishlist

hzhou · 2021-11-22T18:37:01Z

Added a bullet onto the ABI wiki: https://github.com/pmodels/mpich/wiki/ABI-Change-Wishlist

Good point that this is an ABI change, and at least for now, keeping it backward compatible is probably more significant than fixing a warning. That leaves option 1. I guess we can add -Wno-stringop-overflow to configure and add a note in README to advise application to do the same if they desire to use -Werror.

In fact, the more I look at this, the more I'm convinced this is a gcc regression.

hzhou · 2021-11-29T16:52:39Z

@raffenet I am tempted to label this issue as won't fix. It is just warning annoyances. I am also convinced this is really a compiler regression issue. Your opinion?

raffenet · 2021-11-29T17:29:36Z

I agree. My only question is do we submit feedback to the GCC folks?

hzhou · 2021-11-29T17:33:14Z

I agree. My only question is do we submit feedback to the GCC folks?

It will be nice to do that. @travisk1154 Are you interested in submitting the ticket to GCC?

tkoehring · 2021-11-30T21:24:36Z

Sure thing!

hzhou · 2022-01-06T23:01:57Z

Apparently gcc now is making a difference between

int MPI_Waitall(int count, MPI_Request array_of_requests[], MPI_Status array_of_statuses[]);

and

int MPI_Waitall(int count, MPI_Request array_of_requests[], MPI_Status *array_of_statuses);

hzhou · 2022-01-07T15:36:04Z

@balay Could you verify the patch in #5749 resolves the issue?

balay · 2022-01-07T17:17:47Z

@hzhou Thanks! I've tried manual edit to mpi.h [with this change] - and that works!

See: pmodels/mpich#5687

MPICH defines MPI_STATUSES_IGNORE (a pointer) to 1, which raises warnings w/ gcc. This is a known issue that the MPICH devs are not going to fix. See here: pmodels/mpich#5687 This fix suppresses those issues w/ gcc

@jhendersonHDF

* Address nagfor exceptions stoppage. (#3658) * added cmake ieee flag for nagfor * generalized determining the nag compiler * fixing some misc. NAG warnings * Simplify. (#3659) * Address @jhendersonHDF review * Add expedited testing support to t_filters_parallel (#3665) * Remove clang warnings (#3656) * Fixes test failure for gfortran -O2 and -O3, -fdefault-real-16 (#3662) * added cmake ieee flag for nagfor * fixes gfortran -O2 and -O3, -fdefault-real-16 * fixed sync * updated release notes * Fix link error on clang17/gfortran13/macOS-13 (#3666) (#3671) * Correct fortran CMake generator expressions (#3670) * Add AOCC GitHub Action (#3504) (#3657) * Fix uninitialized subfiling test variable (#3675) Picked up by gcc 10 on skybridge. Probably spurious, but no harm in initializing it to a "bad" value. * Add support for AOCC & Flang w/ the Autotools (#3674) * Adds a config/clang-fflags options file to support Flang * Corrects missing "-Wl," from linker options in the libtool wrappers when using Flang, the MPI Fortran compiler wrappers, and building the shared library. This would often result in unrecognized options like -soname. * Enable -nomp w/ Flang to avoid linking to the OpenMPI library. CMake can build the parallel, shared library w/ Fortran using AOCC and Flang, so no changes were needed for that build system. Fixes GitHub issues #3439, #1588, #366, #280 * Fix a strncpy call to use dest size not src (#3677) A strncpy call in a path construction call used the size of the src buffer instead of the dest buffer as the limit n. This was switched to use the dest size and properly terminate the string if truncation occurs. * Remove CANBE_UNUSED() from subfiling VFD (#3678) This macro was an attempt to quiet warnings about release mode unused variables that only appear in asserts. It resolves to a void cast, which doesn't quiet warnings when an assignment has already taken place. * Suppress MPI_Waitall warnings w/ MPICH (#3680) MPICH defines MPI_STATUSES_IGNORE (a pointer) to 1, which raises warnings w/ gcc. This is a known issue that the MPICH devs are not going to fix. See here: pmodels/mpich#5687 This fix suppresses those issues w/ gcc * Fix a possible NULL pointer dereference in tests (#3676) The dtypes test could dereference a NULL pointer if a strdup call failed. * Fix printf warnings in t_mpi (#3679) * Fix printf warnings in t_mpi The type of MPI_Offset varies with implementation. In MPICH, it's long, which raises warnings when we attempt to use long long format specifiers. Casting to long long fixes the warnings. * Fix invalid memory access in S3 comms (#3681) In the ros3 VFD, passing an empty string parameter to an internal API call could result in accessing the -1th element of a string. This would cause failures on big-endian systems like s390x. This parameter is now checked before writing to the string. Fixes GitHub #1168 * Add Doxygen for H5Pset_fapl_sec2() (#3685) * * switch to using time function instead of date function (#3690) * Initialize API context MPI types to MPI_BYTE (#3688) * Add test info output to t_filters_parallel (#3696) * Suppress format string warnings in subfiling test (#3699) * Fix unused variable in tselect.c (#3701) * Fix unused variable warning in H5F_sfile_assert_num (#3700) * Restore floating-point suffixes in tests (#3698) A prior commit removed too many F suffixes. This restores the suffixes for float variables. * Sync with changes from develop --------- Co-authored-by: Scot Breitenfeld <brtnfld@hdfgroup.org> Co-authored-by: H. Joe Lee <hyoklee@hdfgroup.org> Co-authored-by: Allen Byrne <50328838+byrnHDF@users.noreply.github.com> Co-authored-by: Dana Robinson <43805+derobins@users.noreply.github.com>

balay changed the title ~~Compile warnings with MPI_STATUSES_IGNORE~~ [32bit] Compile warnings with MPI_STATUSES_IGNORE, MPI_STATUS_IGNORE Nov 19, 2021

balay changed the title ~~[32bit] Compile warnings with MPI_STATUSES_IGNORE, MPI_STATUS_IGNORE~~ Compile warnings with MPI_STATUSES_IGNORE, MPI_STATUS_IGNORE with gcc-11.2.1 on Fedora-35 Nov 20, 2021

hzhou changed the title ~~Compile warnings with MPI_STATUSES_IGNORE, MPI_STATUS_IGNORE with gcc-11.2.1 on Fedora-35~~ misc: warnings on MPI_STATUS_IGNORE etc with gcc-11.2.1 Nov 29, 2021

hzhou assigned tkoehring Nov 29, 2021

hzhou mentioned this issue Jan 6, 2022

mpi: use pointers for array_of_statuses #5749

Merged

4 tasks

hzhou unassigned tkoehring Jan 6, 2022

hzhou closed this as completed in #5749 Jan 7, 2022

holke added a commit to DLR-AMR/t8code that referenced this issue Jun 3, 2022

Disable stringop-overflow GCC warning temporarily.

ef3eddf

See: pmodels/mpich#5687

holke added a commit to DLR-AMR/t8code that referenced this issue Jun 3, 2022

Disable stringop-overflow GCC warning temporarily.

2fcf081

See: pmodels/mpich#5687

holke added a commit to DLR-AMR/t8code that referenced this issue Jun 3, 2022

Disable stringop-overflow GCC warning temporarily.

c67d9f5

See: pmodels/mpich#5687

thomasgillis mentioned this issue Jan 17, 2023

build: warning with gcc-11.2.0 for MPIC_Waitall #6366

Closed

derobins mentioned this issue Oct 15, 2023

Suppress MPI_Waitall warnings w/ MPICH HDFGroup/hdf5#3680

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

misc: warnings on MPI_STATUS_IGNORE etc with gcc-11.2.1 #5687

misc: warnings on MPI_STATUS_IGNORE etc with gcc-11.2.1 #5687

balay commented Nov 19, 2021

balay commented Nov 20, 2021

hzhou commented Nov 22, 2021 •

edited

Loading

balay commented Nov 22, 2021

hzhou commented Nov 22, 2021

raffenet commented Nov 22, 2021

raffenet commented Nov 22, 2021

hzhou commented Nov 22, 2021 •

edited

Loading

hzhou commented Nov 29, 2021

raffenet commented Nov 29, 2021

hzhou commented Nov 29, 2021

tkoehring commented Nov 30, 2021

hzhou commented Jan 6, 2022

hzhou commented Jan 7, 2022

balay commented Jan 7, 2022

misc: warnings on MPI_STATUS_IGNORE etc with gcc-11.2.1 #5687

misc: warnings on MPI_STATUS_IGNORE etc with gcc-11.2.1 #5687

Comments

balay commented Nov 19, 2021

balay commented Nov 20, 2021

hzhou commented Nov 22, 2021 • edited Loading

balay commented Nov 22, 2021

hzhou commented Nov 22, 2021

raffenet commented Nov 22, 2021

raffenet commented Nov 22, 2021

hzhou commented Nov 22, 2021 • edited Loading

hzhou commented Nov 29, 2021

raffenet commented Nov 29, 2021

hzhou commented Nov 29, 2021

tkoehring commented Nov 30, 2021

hzhou commented Jan 6, 2022

hzhou commented Jan 7, 2022

balay commented Jan 7, 2022

hzhou commented Nov 22, 2021 •

edited

Loading

hzhou commented Nov 22, 2021 •

edited

Loading