[llvm] [AMDGPU] Handle hazard in v_cvt_scalef32_pk_{fp|bf}8_{f|bf}16. (PR #140218)

[Pravin Jagtap via llvm-commits]

pravinjagtap wrote:

This is the cleanest way to handle this. I thought of adding searchable table instead of hardcoding the opcodes. Filtering opcodes from cvt_scale family of instructions based on opsel values and/or types it really difficult in this case. I understand, this api has become really complicated because of custom requirements. 

https://github.com/llvm/llvm-project/pull/140218