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

[via llvm-commits]

================
@@ -1744,6 +1744,12 @@ Value *SplitPtrStructs::handleMemoryInst(Instruction *I, Value *Arg, Value *Ptr,
     case AtomicRMWInst::FMin:
       IID = Intrinsic::amdgcn_raw_ptr_buffer_atomic_fmin;
       break;
+    case AtomicRMWInst::USubCond:
+      IID = Intrinsic::amdgcn_raw_ptr_buffer_atomic_usub_cond;
+      break;
+    case AtomicRMWInst::USubSat:
+      IID = Intrinsic::amdgcn_raw_ptr_buffer_atomic_usub_sat;
+      break;
----------------
anjenner wrote:

I have added buffer-fat-pointer-atomicrmw-usub_cond.ll and buffer-fat-pointer-atomicrmw-usub_sat.ll tests in the latest version of this patch. I needed to implement more pieces of buffer fat pointer support for usub_cond and usub_sat to get these tests to pass.

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