Fix some more uses of LLVM intrinsics

[sayantn]

• Use rust intrinsics for more ARM intrinsics
• Fix incorrect intrinsic names (these were autoupgraded by LLVM, so didn't result in invalid IR)
• Remove uses of deprecated type-specific pointers, LLVM only supports opaque pointers now.

@folkertdev is it okay to replace the llvm.{u,s}{max,min} calls with something like simd_select(simd_gt(a, b), a, b))? I am just worried if it will optimize well

edit: seems like it is not a problem actually https://godbolt.org/z/6r1javWc6

[rustbot]

r? @Amanieu

rustbot has assigned @Amanieu.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[folkertdev]

Just try it, you can check whether that works by running the docker container. We should be testing for the specific instruction (adjust for your target):

TARGET=aarch64-unknown-linux-gnu sh ci/run-docker.sh aarch64-unknown-linux-gnu

If it doesn't work, let me know because we should get that fixed in LLVM.

[Amanieu]

@folkertdev is it okay to replace the llvm.{u,s}{max,min} calls with something like simd_select(simd_gt(a, b), a, b))? I am just worried if it will optimize well

No, those 2 are not equivalent, specifically with regards to negative zero and NaN.

[folkertdev]

umax and smax are only for integers right? For floats the expressions are indeed not equivalent in some nasty ways.

[Amanieu]

The fact that LLVM generates the same code for both suggests a bug in the LLVM backend.

Nevermind, it's fine if this is only for integers.