Feature detection issue on x86 targets

[landonxjames]

I've recently integrated the crc64fast-nvme crate with the aws-sdk-rust to support Crc64 checksums for S3. Everything works as expected except the crate fails to compile on x86. You can see an example failure from our CI here.

error[E0425]: cannot find function `_mm_extract_epi64` in this scope
    --> /cargo/registry/src/index.crates.io-6f17d22bba15001f/crc64fast-nvme-1.0.0/src/pclmulqdq/x86.rs:53:9
     |
53   |         _mm_extract_epi64(reduced.0, 1) as u64
     |         ^^^^^^^^^^^^^^^^^ help: a function with a similar name exists: `_mm_extract_epi16`
     |
    ::: /rust/lib/rustlib/src/rust/library/core/src/../../stdarch/crates/core_arch/src/x86/sse2.rs:1417:1
     |
1417 | pub unsafe fn _mm_extract_epi16<const IMM8: i32>(a: __m128i) -> i32 {
     | ------------------------------------------------------------------- similarly named function `_mm_extract_epi16` defined here

The compiler fails to find the _mm_extract_epi64 function which is present on x86_64, but not on 32 bit x86. I can see in the code that there is already a check for this particular intrinsic:

crc64fast-nvme/src/pclmulqdq/x86.rs

Line 19 in 2450b26

&& is_x86_feature_detected!("sse4.1") // _mm_extract_epi64

so I expect that this case was meant to be covered, but isn't quite working as expected.

[onethumb]

@landonxjames I guess this is what I get for developing on Apple Silicon and deploying to Graviton. 😆 🤦‍♂️

I'll try to find some time to investigate, but feel free to submit a PR if you have a fix in mind. 👍

[onethumb]

@landonxjames I believe this should be fixed in 1.0.1. Crates.io has also been updated.

Let me know if this doesn't do the trick or I missed something. 👍

[onethumb]

@landonxjames Totally unrelated, but love your thoughts on the Double throughput on systems which support AVX512 VPCLMULQDQ PR too, if you get a chance.

[onethumb]

@landonxjames You might be interested to see the new C-compatible shared library support, and an implementation in PHP using FFI. This is released in 1.1.0.

Can you verify the x86 issue is resolved so I can close this Issue?

[landonxjames]

Sorry, was out for the holidays, but I just bumped the version and reran our CI and all looks good to me. Thanks for the fix!