Test that target feature mix up with homogeneous floats is sound #101381
Conversation
|
(rust-highfive has picked a reviewer for you, use r? to override) |
|
r? @Amanieu perhaps? I'm not a huge fan of just running binaries without checking CPUID ahead of time for the relevant target features. I think our CI runs on fairly old CPUs (Intel(R) Xeon(R) CPU E5-2673 v4 @ 2.30GHz, in one recent job), which don't have any AVX-512 features, so we'd not be testing this regularly either. |
|
Fundamentally this LGTM. If we absolutely want to make sure that the AVX512 path is tested then we may want to run the test under |
Is is required ? Personally I don't it should be, qemu is not perfect and most contributors will run those tests on their machines, so it will get tested as much but still enough to get me to want to integrate it. Having the test is better than not having it at all. |
|
No, I don't think this is strictly required. This test is fine as it is. @bors r+ |
|
⌛ Testing commit 5c6caf3 with merge 97c338f2fa6662ed23b3cd8d295a529a8a14487f... |
|
💔 Test failed - checks-actions |
This comment has been minimized.
This comment has been minimized.
…ts, r=Amanieu Test that target feature mix up with homogeneous floats is sound This pull-request adds a test in `src/test/abi/` that test that target feature mix up with homogeneous floats is sound. This is basically is ripoff of [src/test/ui/simd/target-feature-mixup.rs](https://github.com/rust-lang/rust/blob/47d1cdb0bcac8e417071ce1929d261efe2399ae2/src/test/ui/simd/target-feature-mixup.rs) but for floats and without `#[repr(simd)]`. *Extracted from rust-lang#97559 since I don't yet know what to do with that PR.*
|
Wait, never mind. This one failed in CI and wasn't fixed. I wonder how it got back into the bors queue? |
|
@bors r- |
Not the first time. It sometimes happens that the queue gets de-synchronized. Anyway, I looked a bit at the failure and it happen when testing the first It's maybe related to sse not being available on CI but it would be weird to not get a SIGGILL in that case. Anyone know if the builder |
|
The |
|
I tried pretty hard to reproduce the crash (with/without qemu, with/without mold, with/without native libs, ...) but I'm still unable to reproduce it. It always successfully pass. I'm not sure what to do or try anymore. Does anyone have an idea? |
|
Can you reproduce the issue with |
Seems like yes, which does not really make sense to me. Anyway this is progress. I will try to investigate this week-end otherwise next week. |
This is basically is ripoff of src/test/ui/simd/target-feature-mixup.rs but for floats and without #[repr(simd)]
5c6caf3
to
66847ff
Compare
|
I've been unable to find the exact cause of the problem but since @rustbot ready |
|
@bors r+ |
…earth Rollup of 7 pull requests Successful merges: - rust-lang#100508 (avoid making substs of type aliases late bound when used as fn args) - rust-lang#101381 (Test that target feature mix up with homogeneous floats is sound) - rust-lang#103353 (Fix Access Violation when using lld & ThinLTO on windows-msvc) - rust-lang#103521 (Avoid possible infinite loop when next_point reaching the end of file) - rust-lang#103559 (first move on a nested span_label) - rust-lang#103778 (Update several crates for improved support of the new targets) - rust-lang#103827 (Properly remap and check for substs compatibility in `confirm_impl_trait_in_trait_candidate`) Failed merges: r? `@ghost` `@rustbot` modify labels: rollup
This pull-request adds a test in
src/test/abi/that test that target feature mix up with homogeneous floats is sound.This is basically is ripoff of src/test/ui/simd/target-feature-mixup.rs but for floats and without
#[repr(simd)].Extracted from #97559 since I don't yet know what to do with that PR.