Can SIGILL if compile with cpu features and execute without it

[zzau13]

At b_escape function can produce SIGILL cpu exception.

If you compile in to target_arch x86_64 and target_feature avx2 and sse2, you will be forced to execute this function in a cpu with the previous characteristics. In any other way it will produce this exception

There is no cheap way to implement runtime detection until rust-lang/rust#57775 is resolved.

[Chaostheorie]

So this might be a bit late but wouldn't it be possible to add a check for the function, i.e. a check that ensures the relevant input is additionally checked at runtime even if possibly expensive, with a build-time check for the current CPU features (Rust Docs)? This will make yarte slower on CPUs not supporting the features though this solution wouldn't exclude them at least.

[zzau13]

@Chaostheorie Sorry for taking time to respond. Yes. A once call function can be added for check runtime cpu flags are the same as the compile cpu flags. This function could be launched at the beginning of the application and panic in case of error.

[zzau13]

This will make yarte slower on CPUs not supporting the features though this solution wouldn't exclude them at least.

The other architectures are not excluded, for architectures other than x86_64 has no effect, the fallback is launched, basically because it is not implemented with any other architecture.

[paberr]

I've been trying to compile a version of this crate that is suitable for all x86-64 cpus for the use in Docker (i.e., one that does not depend on CPU extensions).
However, setting build targets has no effect because this library heavily uses runtime feature checks – even during the build process.

For example, in the build.rs, it will check the CPU features at compile time instead of the target features.

Is there a reason why the library relies so much on is_x86_feature_detected!("avx2") instead of cfg!(target_feature = "avx2") or even #[cfg(target_feature = "avx2")]?