[llvm] [AMDGPU] Teach SIPreEmitPeephole pass to preserve MachineLoopInfo (PR #178868)

[Christudasan Devadasan via llvm-commits]

================
@@ -709,9 +716,14 @@ llvm::SIPreEmitPeepholePass::run(MachineFunction &MF,
                                  MachineFunctionAnalysisManager &MFAM) {
   auto *MDT = MFAM.getCachedResult<MachineDominatorTreeAnalysis>(MF);
   auto *MPDT = MFAM.getCachedResult<MachinePostDominatorTreeAnalysis>(MF);
+  SIPreEmitPeephole Impl;
 
-  if (SIPreEmitPeephole().run(MF))
-    return getMachineFunctionPassPreservedAnalyses();
+  if (Impl.run(MF)) {
+    auto PA = getMachineFunctionPassPreservedAnalyses();
+    if (!Impl.isCFGModified())
----------------
cdevadas wrote:

Yep, don't use this separate checker for CFGModified transformations. It adds unwanted overhead and will become hard to maintain in the future.

https://github.com/llvm/llvm-project/pull/178868