Skip to content

Enable testing on Fedora Rawhide#3024

Open
bkmgit wants to merge 1 commit into
google:masterfrom
bkmgit:packit
Open

Enable testing on Fedora Rawhide#3024
bkmgit wants to merge 1 commit into
google:masterfrom
bkmgit:packit

Conversation

@bkmgit
Copy link
Copy Markdown
Contributor

@bkmgit bkmgit commented Apr 30, 2026

Uses Fedora infrastructure https://packit.dev

Closes #3022

@bkmgit
Copy link
Copy Markdown
Contributor Author

bkmgit commented Apr 30, 2026

At present tests on ppc64 and build on aarch64 fail:
https://koji.fedoraproject.org/koji/taskinfo?taskID=144992149

@bkmgit
Copy link
Copy Markdown
Contributor Author

bkmgit commented Apr 30, 2026

For tests on Fedora to run, enable Packit integration. See:
https://packit.dev/docs/guide#github

Copy link
Copy Markdown
Member

@jan-wassenberg jan-wassenberg left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice. Probably good to wait a bit until we can remove CC0.

@jan-wassenberg
Copy link
Copy Markdown
Member

Had a quick look at the failing builds.

aarch64 Error: unknown pseudo-op: `.lbe73430'
Probably OOM leading to truncated assembly file. I wonder if we can influence the flags, e.g. remove -ffat-lto-objects?
Might also help to split compare_test and/or disable some targets, I can do that.

ppc:

[----------] 1 test from HwyMemoryTestGroup/HwyMemoryTest
[ RUN      ] HwyMemoryTestGroup/HwyMemoryTest.TestAllLoadDup128/EMU128
i16x8 expect [0+ ->]:
  0x0001,0x0002,0x0003,0x0004,0x0005,0x0006,0x0007,
i16x8 actual [0+ ->]:
  0x0110,0x0000,0x0000,0x0000,0x0001,0x0002,0x0003,
Abort at memory_test.cc:393221: EMU128, i16x8 lane 0 mismatch: expected '0x0001', got '0x0110'.

This kind of thing is usually a compiler mistake caused by incorrect constant folding. I'm adding hwy::Unpredictable1() there which might help.

copybara-service Bot pushed a commit that referenced this pull request Apr 30, 2026
Add Unpredictable1 instead of literal

Also disable some aarch64 targets for compare_test to prevent OOM

PiperOrigin-RevId: 908194619
@bkmgit
Copy link
Copy Markdown
Contributor Author

bkmgit commented May 1, 2026

For aarch64 another issue is fixed by #3030
If fewer cores are used to compile, the build succeeds.

Failing build that resulted in #3030
https://koji.fedoraproject.org/koji/taskinfo?taskID=145011276
Build with fix:
https://koji.fedoraproject.org/koji/taskinfo?taskID=145019086

Disabling LTO also reduces memory requirements, but not using fat LTO objects might be sufficient in
most cases, see:
https://stackoverflow.com/questions/13799452/what-is-the-difference-in-gcc-between-lto-and-fat-lto-objects

@bkmgit
Copy link
Copy Markdown
Contributor Author

bkmgit commented May 2, 2026

AArch64 on Copr seems to have another failing test:
https://copr.fedorainfracloud.org/coprs/fed500/highway/build/10415978/

The following tests FAILED:
1546 - HwyReductionTestGroup/HwyReductionTest.TestAllMaskedReduceSum/SVE_256 (Subprocess aborted)
1558 - HwyReductionTestGroup/HwyReductionTest.TestAllMaskedReduceMax/SVE_256 (Subprocess aborted)

The machines used on Copr are likely different than the ones used on the koji tests. It may be helpful to log more information about the CPU microarchitecture being used and available RAM when doing tests.

Arm seems to have several other bugs:
#2974

copybara-service Bot pushed a commit that referenced this pull request May 4, 2026
Add Unpredictable1 instead of literal

Also disable some aarch64 targets for compare_test to prevent OOM

PiperOrigin-RevId: 908194619
copybara-service Bot pushed a commit that referenced this pull request May 4, 2026
Add Unpredictable1 instead of literal

Also disable some aarch64 targets for compare_test to prevent OOM

PiperOrigin-RevId: 909853992
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Test on leading edge linux distribution

2 participants