Steam hardware survey has support at >99% so I think this is fine. If the CI fails again I'll grab and test this on mac myself. Thanks!

(I think bevy is using this repo so I asked in their discord for anyone with objections to speak up.)

It seems preferable to check whether sse4.1 is listed in CARGO_CFG_TARGET_FEATURE. That will indicate whether the user requested rustc to produce binaries containing SSE 4.1 instructions, rather than simply inferring it by what ISA is used on the build machine.

Someone please correct me if I get this wrong but the tier 1 targets in rust are x86_64, i686, and aarch64. You can check features with something like rustc --print=cfg -C target-cpu=x86-64. On windows it looks like this

debug_assertions
panic="unwind"
target_arch="x86_64"
target_endian="little"
target_env="msvc"
target_family="windows"
target_feature="fxsr"
target_feature="sse"
target_feature="sse2"
target_has_atomic="16"
target_has_atomic="32"
target_has_atomic="64"
target_has_atomic="8"
target_has_atomic="ptr"
target_os="windows"
target_pointer_width="64"
target_vendor="pc"
windows

None of these tier 1 targets include sse4.1. So while checking the target does seem like the intended solution, in practice I think this will mean almost no one will enable sse4.1. (I expect most people take the default target for their OS.)

This is not to say I object to going with this, but I think it's worth mentioning.

The default target-cpu for each of the tier 1 targets doesn't include sse4.1. However, you can set a different target-cpu via RUSTFLAGS. You can see the list of supported CPUs to pick from with something like:

$ rustc --target=x86_64-unknown-linux-gnu --print target-cpus

The list includes a couple of "x86-64 microarchitecture levels" in addition to named cpu models. Actually using it would look something like:

$ RUSTFLAGS='-C target-cpu=x86-64-v2' cargo run

(You can instead create a .cargo/config in the root of your repository if you don't want to write it out every time)

Based on the input so far, going to hold off on taking this PR until it is opt-in, preferably by checking CARGO_CFG_TARGET_FEATURE

Thanks for all your input!

I think letting the user pass compiler flags is too arcane, so I exposed it as a regular feature hint_sse4_1 + an additional check if we are building for x86 or x86_64. So, if the hint_sse4_1 flag is set and where building for non-x86, no SEE 4.1 is used. That should prevent headaches when building for non x86 platforms.