Add rsqrt method to Float trait

[davll]

rsqrt is a widely used math function in game development, and is faster than combining the two functions: recip(sqrt(x)) thanks to x86 SSE instructions RSQRTSS, RSQRTPS, RSQRTSD, and RSQRTPD. Should we consider add rsqrt to Float trait?

• x86/x86_64 SSE: _mm_rsqrt_ps (note that RSQRT instruction is approximate, less accurate than SQRT)
• ARM Neon VRSQRTE
• PowerPC Altivec: vec_rsqrte

[cuviper]

This should probably be proposed for the standard library first. So far we don't have any arch-specific code -- I'd rather just have num's impl Float for f64 forward to an optimized/intrinsic call in std.

[clarfonthey]

I agree; it makes sense to offer this in libstd.

[davll]

Reasonable, I'll propose it in rust-lang/rust.
I'll keep the issue open as reminder.

[cuviper]

This issue was moved to rust-num/num-traits#1