[llvm] AMDGPU: Fix s_barrier_leave to write to scc (PR #161221)

[Jay Foad via llvm-commits]

================
@@ -1710,6 +1710,8 @@ def VOP3PMadMixBF16Mods : ComplexPattern<untyped, 2, "SelectVOP3PMadMixBF16Mods"
 def VINTERPMods  : ComplexPattern<untyped, 2, "SelectVINTERPMods">;
 def VINTERPModsHi  : ComplexPattern<untyped, 2, "SelectVINTERPModsHi">;
 
+def Ignore : ComplexPattern<untyped, 0, "SelectIgnore">;
----------------
jayfoad wrote:

@s-barannikov @jurahul @Pierre-vh FYI - this patch introduces this dummy complex pattern as a way to let you ignore an operand in an isel pattern. I wonder if tablegen should have some kind of built-in support for this instead?

https://github.com/llvm/llvm-project/pull/161221