[llvm] [AMDGPU][True16][CodeGen] Support AND/OR/XOR and LDEXP True16 format (PR #102620)

[Brox Chen via llvm-commits]

================
@@ -236,5 +243,38 @@ bool GCNPreRAOptimizations::runOnMachineFunction(MachineFunction &MF) {
     Changed |= processReg(Reg);
   }
 
+  if (!ST.useRealTrue16Insts())
+    return Changed;
+
+  // Add RA hints to improve True16 COPY elimination.
----------------
broxigarchen wrote:

I remember we have a plan to do the coalescing altogether in a later time.  @Sisyph would know more about it.  Hi Joe, do you have any comment on this?

https://github.com/llvm/llvm-project/pull/102620