#intel #sse #avx

nightly immintrin

Implementation of Intel x86 intrinsics (_mm_*)

2 unstable releases

Uses old Rust 2015

0.2.0 Jul 31, 2016
0.1.0 Jul 12, 2016

#11 in #avx

MIT/Apache

84KB
2.5K SLoC

immintrin

This crate implements the names defined by Intel for SIMD and related instructions on x86 (functions with an _mm_ prefix). These are useful for porting code, and because people are familiar with them. See https://github.com/rust-lang/rfcs/issues/1639 for discussion.

Currently contains signatures for all SSE/SSE2/SSE2/SSSE3 intrinsics. Intrinsics involving MMX and non-temporal stores are not yet implemented; everything else should work. Requires a nightly compiler for SIMD and various unstable intrinsics.

If anyone is looking to improve this crate, probably the most important thing at the moment is some sort of testing infrastructure; currently, this crate has no tests. Not precisely sure what the tests would look like; probably some combination of checking the generated LLVM IR, checking the generated assembly, and runtime tests.

It's not clear whether the signatures for intrinsics which take a constant integer should stay the way they are, or be changed somehow. Currently, we just accept that the parameter might not be constant, and just make sure we generate code which will be optimized well if the parameter is constant (for example, the match statement in _mm_slli_si128 will fold down to a single instruction).

Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0> or the MIT license <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your option. All files in the project carrying such notice may not be copied, modified, or distributed except according to those terms.

Dependencies

~155KB