simd-accel no longer works (encoding_rs's SIMD support is broken on the latest nightly)

[lilydjwg]

What version of ripgrep are you using?

N/A.

How did you install ripgrep?

Failed.

What operating system are you using ripgrep on?

I'm trying to build on Arch Linux with Intel(R) Xeon(R) CPU E5-2620 v3 and I'm going to run it on Arch Linux with Intel(R) Core(TM) i7-7700HQ.

If this is a bug, what are the steps to reproduce the behavior?

I'm building with:

RUSTFLAGS="-C target-cpu=skylake" cargo build --release --features 'simd-accel pcre2'

If this is a bug, what is the actual behavior?

I got a lot of errors like this:

error: unrecognized platform-specific intrinsic function: `x86_mm_movemask_ps`      
  --> /build/.cargo/registry/src/github.com-1ecc6299db9ec823/simd-0.2.3/src/x86/sse2.rs:10:5
(more omitted)

If this is a bug, what is the expected behavior?

There shouldn't be such errors.

---

It seems that cargo is trying to build the build script with options intended for the target system.

rustc and cargo are nightly verions.

[BurntSushi]

This has nothing to do with whether the build system supports the prescribed target features. The issue is that encoding_rs's SIMD support is now broken on the latest release of nightly Rust. The only way it becomes unbroken is if nightly Rust reverts the change removing unstable legacy support for for platform specific intrinsics, or if encoding_rs migrates to the new SIMD infrastructure. The latest is, AFAIK, tracked by hsivonen/encoding_rs#23.

I've updated the README in 9a9f54d to note that ripgrep's simd-accel feature is broken.

[lilydjwg]

I see, thank you.

[hsivonen]

It works now if the lock file is updated with cargo update first. The README for encoding_rs has the details of the pitfalls of trying to build for 32-bit ARM targets whose first component does not have the substring neon.

[BurntSushi]

@hsivonen Awesome! Thanks so much. Cargo.lock and README have been updated!