[llvm] MachineScheduler: Improve instruction clustering (PR #137784)

[Jay Foad via llvm-commits]

jayfoad wrote:

> For AMDGPU, some cases have more v_dual instructions used while some are regressed. It seems less critical.

Agreed, this is probably just due to good luck or bad luck in register allocation. We could put `-amdgpu-enable-vopd=0` on these tests.

https://github.com/llvm/llvm-project/pull/137784