[llvm] [AMDGPU][GFX12] Add Atomic cond_sub_u32 (PR #76224)

[Matt Arsenault via llvm-commits]

================
@@ -2491,6 +2499,11 @@ def int_amdgcn_flat_atomic_fmax_num   : AMDGPUAtomicRtn<llvm_anyfloat_ty>;
 def int_amdgcn_global_atomic_fmin_num : AMDGPUAtomicRtn<llvm_anyfloat_ty>;
 def int_amdgcn_global_atomic_fmax_num : AMDGPUAtomicRtn<llvm_anyfloat_ty>;
 
+def int_amdgcn_flat_atomic_cond_sub_u32 : AMDGPUAtomicRtn<llvm_i32_ty, flat_ptr_ty>;
+def int_amdgcn_global_atomic_cond_sub_u32 : AMDGPUAtomicRtn<llvm_i32_ty, global_ptr_ty>;
+
+def int_amdgcn_ds_cond_sub_u32 : AMDGPUAtomicRtn<llvm_i32_ty, local_ptr_ty>;
----------------
arsenm wrote:

The other memory operations we just have one intrinsic mangled by the pointer type. We should also consider just adding it to atomicrmw 

https://github.com/llvm/llvm-project/pull/76224