[llvm] [X86] computeKnownBitsForTargetNode - add INTRINSIC_WO_CHAIN handling for PSADBW intrinsics (PR #83580)

[Phoebe Wang via llvm-commits]

https://github.com/phoebewang commented:

> Waiting for intrinsics to be lowered to ISD target nodes is causing some poor combine decisions - at the very least we need better value tracking handling.

I don't quite understand the point. Shouldn't the first combine happens after first round lowering?

https://github.com/llvm/llvm-project/pull/83580