redhat gcc 14 ICE on i386

[mr-c]

FAILED: test/arm/neon/ext-native-c.p/ext.c.o 
gcc -Itest/arm/neon/ext-native-c.p -Itest/arm/neon -I../../test/arm/neon -I. -I../.. -fdiagnostics-color=always -pipe -D_FILE_OFFSET_BITS=64 -Wall -Winvalid-pch -std=c99 -g -O2 -fopenmp-simd -DSIMDE_CONSTRAINED_COMPILATION -DSIMDE_ENABLE_OPENMP -Wno-psabi -DSIMDE_TEST_BARE -ffloat-store -MD -MQ test/arm/neon/ext-native-c.p/ext.c.o -MF test/arm/neon/ext-native-c.p/ext.c.o.d -o test/arm/neon/ext-native-c.p/ext.c.o -c ../../test/arm/neon/ext.c
during GIMPLE pass: local-fnsummary
../../test/arm/neon/ext.c: In function ‘simde_vext_p64’:
../../test/arm/neon/ext.c:2236: internal compiler error: in predicate_for_phi_result, at ipa-fnsummary.cc:2445
Please submit a full bug report, with preprocessed source (by using -freport-bug).
See <http://bugzilla.redhat.com/bugzilla> for instructions.
Preprocessed source stored into /tmp/cchL3c7C.out file, please attach this to your bugreport.

https://download.copr.fedorainfracloud.org/results/packit/simd-everywhere-simde-1127/fedora-rawhide-i386/06994644-simde/builder-live.log.gz https://download.copr.fedorainfracloud.org/results/packit/simd-everywhere-simde-1127/fedora-rawhide-i386/06994644-simde/ https://dashboard.packit.dev/results/copr-builds/1313878

@junaruga can you file the bug with RedHat?

Also, on s390x we get new test failures:

17:55:09 /builddir/build/BUILD/simde-v0.8.0.rc1/build/gcc-O2/test/arm/neon/addv-native-c
not ok 8 addv/vaddvq_f32
../../test/arm/neon/addv.c:357: assertion failed: r ~= test_vec[i].r (0.000000 ~= -1780.969971)

(same for the neon/addv emul and C++ variants) https://download.copr.fedorainfracloud.org/results/packit/simd-everywhere-simde-1127/fedora-rawhide-s390x/06994644-simde/builder-live.log.gz

[junaruga]

Do you know which commit in this repository triggered the failure? Or a new version gcc 14 on Fedora rawhide triggered the failure? If you think it is a bug of the gcc14, one possible way to report the issue to the upstream GCC project.

You know the detail of the bug. You can create an account, and report the issue with selecting component: gcc on Bugzilla.

[junaruga]

Sorry, I am negative the issue to the Fedora project by myself.

[junaruga]

I think reporting a bug to the GCC project is a typical use case in this project seeing the past cases.

• https://github.com/simd-everywhere/simde/issues/124#issuecomment-606282818
• https://github.com/simd-everywhere/simde/issues/949
• https://github.com/simd-everywhere/simde/issues/674#issuecomment-806046596

[junaruga]

If you want to reproduce the issue on Fedora rawhide on your local, below are the container images.

• https://quay.io/repository/fedora/fedora
• https://hub.docker.com/_/fedora/

[mr-c]

@junaruga Yeah, we don't do any gcc 14 testing yet besides via your Fedora CI setup; it would be a big help if you had the time to file GCC bugs for us

[Artoria2e5]

Likely unrelated, but your environment has an encoding issue: ‘simde_vext_p64’ indicates that you have UTF-8 output from gcc (presumably a remote machine with LC_ALL=something.UTF-8), but your terminal program is decoding it as either Latin-1 or Windows-1252. You might want to get that fixed; the correct text is ‘simde_vext_p64’.

[sharkcz]

a couple notes

• if access to a s390x machine would help, please let me know
• because I am not familiar with simde at all, is there a way to run the single failing test so I can take a closer look?

[mr-c]

• because I am not familiar with simde at all, is there a way to run the single failing test so I can take a closer look?

Thank you for asking, I've added some documentation about this https://github.com/simd-everywhere/simde/pull/1140/files?short_path=eca12c0#diff-eca12c0a30e25b4b46522ebf89465a03ba72a03f540796c979137931d8f92055 ; please leave a comment there with any feedback you have.

So for the s390x test mentioned above:

meson test --print-errorlogs arm/neon/addv-native-c

[sharkcz]

for the record, we have now https://bugzilla.redhat.com/show_bug.cgi?id=2265489 and https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114075 for the s390x test failures (miscompilations) with gcc 14

[mr-c]

Thank you @sharkcz !! Can you add those bugs to our tracking table at https://github.com/simd-everywhere/simde/wiki/Compiler-Bugs ?

[sharkcz]

for the record, https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114075 was fixed a while ago, so s390x should pass the tests again

[mr-c]

Thank you @sharkcz ; I am seeing that at https://copr.fedorainfracloud.org/coprs/packit/simd-everywhere-simde-1154/build/7211582/

Can you help us by filing the bug for i386?

[sharkcz]

I am going to leave it to @junaruga , he is the simde maintainer in Fedora and my focus is on ppc64le and s390x. It shouldn't be difficult though, as it's an ICE.

[junaruga]

Seeing the copr build page, I see the latest GCC 14.0.1-0.12.fc41 is used. As I don't see the i686 (i386) failure for the build with simde version 0.7.6 on the rawhide, the issue may be this repository's master branch or newer versions specific.

I hope we drop the Packit CI's i386 (i686) case on this repository. I don't think that I want to spend time to work for the i686 any more. Then I will drop the i686 support in simde RPM package on Fedora too. Below are the situation around the i386 (i686) on Fedora project. The i386 (i686) is a kind of deprecated.

• https://fedoraproject.org/wiki/Changes/EncourageI686LeafRemoval
• https://fedoramagazine.org/in-fedora-31-32-bit-i686-is-86ed/

[junaruga]

I hope we drop the Packit CI's i686 (i386) case on this repository. I don't think that I want to spend time to work for the i686 any more. Then I will drop the i686 support in simde RPM package on Fedora too. Below are the situation around the i686 (i386) on Fedora project.

I sent the PR to drop the i386 (i686) case on Packit CI now. https://github.com/simd-everywhere/simde/pull/1155

[mr-c]

Thank you @junaruga !