[Mlir-commits] [llvm] [mlir] Add cond_sub and sub_clamp operations to atomicrmw (PR #96661)

[Matt Arsenault]

================
@@ -132,6 +134,10 @@ Changes to the AMDGPU Backend
 
 * Implemented :ref:`llvm.get.rounding <int_get_rounding>` and :ref:`llvm.set.rounding <int_set_rounding>`
 
+* Removed ``llvm.amdgcn.atomic.cond.sub.u32`` and
----------------
arsenm wrote:

ping, would be best to split all of the AMDGPU implementation pieces into a separate pR 

https://github.com/llvm/llvm-project/pull/96661