[llvm] [AMDGPU] Correctly insert s_nops for implicit read of SDWA (PR #100276)

[Jeffrey Byrnes via llvm-commits]

jrbyrnes wrote:

Adjust condition for "consumer" of forwarded dest. A VOP3A instruction can have partial write to the same dest using op_sel -- in this case, there will be implicit read on initial dest for parity check for ECC. The correctness issue is not isolated then to selecting the high bits via op_sel (as was previously implemented). This only affects ECC and not the actual bits in the dest reg.

https://github.com/llvm/llvm-project/pull/100276