[llvm] [AMDGPU][MC] Fix disassembly for v_permlane16_swap_b32 for GFX950 (PR #146600)

[Joe Nash via llvm-commits]

================
@@ -389,8 +389,8 @@ def VOP_PERMLANE_SWAP : VOPProfile<[i32, i32, untyped, untyped]> {
   let HasExtDPP = 0;
   let HasExtSDWA = 0;
 
-  let Ins32 = (ins Src0RC64:$vdst_in, Src0RC32:$src0);
-  let Ins64 = (ins Src0RC64:$vdst_in, Src0RC64:$src0, Dpp16FI:$fi, DppBoundCtrl:$bound_ctrl);
+  let Ins32 = (ins DstRC:$vdst_in, Src0RC32:$src0);
+  let Ins64 = (ins DstRC64:$vdst_in, Src0RC64:$src0, Dpp16FI:$fi, DppBoundCtrl:$bound_ctrl);
   let InsVOP3OpSel = (ins Src0RC64:$vdst_in, Src0RC64:$src0, Dpp16FI:$fi, DppBoundCtrl:$bound_ctrl);
----------------
Sisyph wrote:

As a side question, is it necessary to have InsVOP3OpSel here? It is not typical to have to override both InsVOP3OpSel and Ins64 in a profile.

https://github.com/llvm/llvm-project/pull/146600