[llvm] [AMDGPU] [PeepholeOpt] Eliminate unnecessary packing in wider f16 vectors for sdwa/opsel-able instruction (PR #137137)

[Krzysztof Drewniak via llvm-commits]

================
@@ -1361,6 +1379,499 @@ bool SIPeepholeSDWALegacy::runOnMachineFunction(MachineFunction &MF) {
   return SIPeepholeSDWA().run(MF);
 }
 
+static bool isSrcDestFP16Bits(MachineInstr *MI, const SIInstrInfo *TII) {
----------------
krzysz00 wrote:

Is there a way to tablegen this switch-case?

(Also, are there relevant things that use the OPSEL scheme and not the SDWA one?)

https://github.com/llvm/llvm-project/pull/137137