[llvm] AMDGPU: Use pattern to select instruction for intrinsic llvm.fptrunc.round (PR #105761)

[Changpeng Fang via llvm-commits]

changpeng wrote:

> I don't like secret opcodes that are not vendor opcodes. If it is a standard named opcodes, I prefer to advertise your achievemnts.

Yes, it should always be good to document when we introduce something that is new.  Now I don't do the rename of the opcode in this patch, it should be a separate patch to document.

By the way, https://llvm.org/docs/GlobalISel/GenericOpcode.html only lists a subset of the opcodes defined in llvm/include/llvm/Target/GenericOpcodes.td

https://github.com/llvm/llvm-project/pull/105761