remove std::simd

[gnzlbg]

Closes #525 .

[gnzlbg]

I've run out of CI capacity so I am posting this here a bit early, its still WIP but most things build already and most tests on most targets pass.

I won't merge this till the ppv crate can be used as a fallback, but right now the only feature missing is the FromBits/IntoBits traits, which are not part of the RFC. My plan is to add those behind a feature flag, and then release the crate.

[lu-zero]

Keeping FromBits and IntoBits would be less painful for our theoretical users (and for us when writing the tests) :)

[gnzlbg]

Keeping FromBits and IntoBits would be less painful for our theoretical users (and for us when writing the tests) :)

Since we are removing all portable vector types from here, it is going to be painful anyways.

For us.. we can just use transmute for the time being, I don't personally think that the gazillion FromBits/IntoBits implementations that are required to make this ergonomic are worth it.

If someone wants to send a PR adding them for the types in the new simd.rs module, I can review that, but the biggest offenders needed these are the arm/aarch64/ppc modules, x86 for example does not really use them.

An alternative is to just abbreviate transmute to ::t(...) or similar, instead of using ::mem::transmute everywhere...

For this to land, the ppv crate will need to implement the FromBits/IntoBits traits first though.

[CryZe]

This has the downside that you can't experiment with portable SIMD on Godbolt anymore, which was very useful. I guess someone should just add the ability to use crates to Godbolt though instead.

[gnzlbg]

@CryZe yeah that was very useful, godbolt supports using libraries for other languages (e.g. C++) so maybe we should open an issue there. Now that godbolt has gained cross-compilation support, this would be very very useful.

[alexcrichton]

@gnzlbg would you be ok merging this for now to get CI green again, but we'll hold off on updating the submodule in rust-lang/rust until you've published?

[gnzlbg]

Done, I might get time to get the From/IntoBits impl ported tomorrow. The submodule update will still take a while but maybe we could start doing that at the end of this week, since that should fix all of the linker errors that people were seeing due to missing definitions of some of the vector functions.