[llvm] AMDGPU: Make v2f64 -> v2f16 conversion Legal only when unsafe fast math is set (PR #134738)

[Changpeng Fang via llvm-commits]

changpeng wrote:

Custom lowering for v264 -> v2f16 on both SelectionDAG and GlobalISel paths now. Fast math flag for fp_round is completely not working and needs additional work beyond this patch.

https://github.com/llvm/llvm-project/pull/134738