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

[Pierre van Houtryve via llvm-commits]

================
@@ -876,7 +876,7 @@ void AMDGPUAtomicOptimizerImpl::optimizeAtomic(Instruction &I,
   BasicBlock *Predecessor = nullptr;
   if (ValDivergent && ScanImpl == ScanOptions::Iterative) {
     // Move terminator from I's block to ComputeEnd block.
-    Instruction *Terminator = EntryBB->getTerminator();
+    BranchInst *Terminator = cast<BranchInst>(EntryBB->getTerminator());
----------------
Pierre-vh wrote:

Added comment + changed var name

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