[llvm] [AMDGPU] Fix crash in InstCombine (PR #179511)

[Alexey Sachkov via llvm-commits]

AlexeySachkov wrote:

> Maybe these cases should fail the verifier

Added check into the verifier in 6a3f4d59505e551b746ed1624ab9133f010d9c2d

Those intrinsics are also exposed through clang built-ins and as a follow-up we should probably add a clang-level diagnostic too:

https://github.com/llvm/llvm-project/blob/209ff8bf06dd12becd79c1ebb01612c021e19f6c/clang/include/clang/Basic/BuiltinsAMDGPU.td#L911

However, from what I can see in [`ROCm/llvm-project:amd/device-libs/ockl/src/base-image-intrinsics.ll`](https://github.com/ROCm/llvm-project/blob/amd-staging/amd/device-libs/ockl/src/base-image-intrinsics.ll) we still use LLVM IR intrinsics directly.

https://github.com/llvm/llvm-project/pull/179511