Unsafe floating point optimizations

[billyb2]

Hi, would it be possible to maybe have a feature gate for unsafe floating point optimizations within this crate? The ease of using SIMD vectors is really nice, and I think having the option of having the crate run even faster (at the expense of precision of course) would be a great addition.

[bitshifter]

What specific optimizations are you referring to?

[billyb2]

Anything that is in std::intrinsics::*_fast

For example:
https://doc.rust-lang.org/std/intrinsics/fn.fadd_fast.html

[bitshifter]

Generally with glam I have focussed on functionality that is in stable rust or have a clear path to stabilisation. These intrinsics are unstable and as far as I know there is no plan to stabilise them on the horizon (https://doc.rust-lang.org/beta/unstable-book/library-features/core-intrinsics.html). In general the question of how to do "fast math" in Rust is unanswered (see rust-lang/rust#21690 which has been around since 2015). This is one of the reasons I built glam around explicit SIMD usage.

At some point perhaps Rust will get fast math support, but it might not be via explicit intrinsics, e.g. it could be a function attributes that allows the compiler optimise a block by disabling IEEE compliance, who knows. Therefore this isn't something I plan to spend time on. I would consider a PR but accepting it would depend on what the maintenance burden looked like.

[billyb2]

Oh okay, I totally understand. I might later work on a PR for this, so I think I'll keep this issue open for now if that's okay.

[bitshifter]

This has been inactive for a while, closing.