[llvm] AMDGPU: Improve codegen for intrinsic llvm.fptrunc.round (PR #104486)

[Jay Foad via llvm-commits]

jayfoad wrote:

> Improve codegen for intrinsic llvm.fptrunc.round

Do you have an example where codegen is improved?

> We name the pseudo instruction as FPTRUNC_ROUND_F32_TO_F16 (for f32 -> f16)

How about calling it `FPTRUNC_ROUND_F16_F32` to match `V_CVT_F16_F32`?

https://github.com/llvm/llvm-project/pull/104486