[llvm] [AMDGPU][AtomicOptimizer] Fix DT update for divergent values with Iterative strategy (PR #87605)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,88 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -mtriple=amdgcn-- -mcpu=gfx908 -passes="amdgpu-atomic-optimizer,verify<domtree>" %s -S -o - | FileCheck %s
+
+; Check we're properly adding an edge from ComputeEnd to the "End" block added by
+; SplitBlockAndInsertIfThen
+;
+; If the edge is not added, domtree verification will fail.
+
----------------
arsenm wrote:

Can you merge the test with the existing ones?

https://github.com/llvm/llvm-project/pull/87605