[llvm] AMDGPU: Add codegen for atomicrmw operations usub_cond and usub_sat (PR #141068)

[Jay Foad via llvm-commits]

================
@@ -1642,6 +1642,12 @@ defm : FlatAtomicPat <"FLAT_ATOMIC_MIN_F64", "atomic_load_fmin_"#as, f64>;
 defm : FlatAtomicPat <"FLAT_ATOMIC_MAX_F64", "atomic_load_fmax_"#as, f64>;
 }
 
+let SubtargetPredicate = isGFX12Plus in {
+  defm : FlatAtomicRtnPat<"FLAT_ATOMIC_COND_SUB_U32", "atomic_load_usub_cond_"  #as, i32 >;
+
+  let OtherPredicates = [HasAtomicCSubNoRtnInsts] in
----------------
jayfoad wrote:

Another wrong use of "CSub" predicate.

https://github.com/llvm/llvm-project/pull/141068