[llvm] [AArch64][SDAG] Add f16 -> i16 rounding NEON conversion intrinsics (PR #155851)

[Kajetan Puchalski via llvm-commits]

mrkajetanp wrote:

> Is the instruction in https://developer.arm.com/architectures/instruction-sets/intrinsics/#q=vcvth_u32_f16 incorrect, and it should be the same as GCC in https://godbolt.org/z/3xjPe7b3z?

Yes, developer.arm.com seems to be showing an incorrect instruction format there.

> Are there U variants to go with the S variants?

There are, I kept it just to the S variants because those are the only ones I've seen overflow issues with. Though I suppose it doesn't hurt to add the U variants too just in case.



https://github.com/llvm/llvm-project/pull/155851