The LLVM intrinsic doesn't support float64x1_t, but the required
instruction is a scalar form (e.g. `frint32x <Dd>, <Dn>`), so we can
implement these using the scalar intrinsic.
Note that Clang does not support these intrinsics, so they aren't
covered by intrinsic-test. Additional validation is included in this
patch to ensure that we're selecting an instruction with the same
behaviour as the corresponding vector form (which all have
intrinsic-tests).
In almost all cases it is preferable to use the stable `asm!` instead of
calling these intrinsics.
This PR removes the following unstable intrinsics:
- `__breakpoint` Clang extension, not part of ACLE.
- `brk`: undocumented
- `_rev*`, `_clz*`, `_rbit*`: use methods on integer types instead
- `__ldrex`, `__strex`: deprecated in ACLE, hard to use correctly
- Register access API: API doesn't match ACLE, better to just use asm
Also considered for deletion, but not included in this PR:
- Barriers: `__isb`, `__dsb`, `__dmb`
- Hints: `__wfi`, `__wfe`, `__sev`, `__sevl`, `__yield`, `__nop`
This involves moving from the ACLE intrinsic definitions (which aren't
available for SVE at this point) to a JSON file. This was derived from
ARM's documentation[^1], and then relicensed under `MIT OR Apache-2.0` for
use in this repository.
[^1]: https://developer.arm.com/architectures/instruction-sets/intrinsics
These intrinsics actually are zeroing, so we should be upfront
that these still return a valid value in reality. At most,
we might in the future want to use a "freezing" semantics here.
In any case, they are definitely not returning MaybeUninit,
or any other possibly-poison value.
