Use LoopRotate PrepareForLTO stage in NPM

[sanwou01]

Summary

The PrepareForLTO stage of LoopRotate tries to avoid unrolling loops
with calls that might be inlined later. See D94232 where this was
introduced.

We didn't catch all occurances of the LoopRotatePass in the New Pass
Manager, so the original regression in astar returned with the pass
manager switch.

[fhahn]

This matches the behavior in the legacy pass manager. LGTM, thanks!

llvm/lib/Passes/PassBuilder.cpp

519

nit: The helper specifically checks for the 'pre-lto' stage, might be good to clarify in the comment, e.g. something like Helper to check if the current compilation phase is preparing for LTO

521

nit: might be slightly simpler to just have return Phase == ThinOrFullLTOPhase::ThinLTOPreLink || Phase == ThinOrFullLTOPhase::ThinLTOPreLink;

[fhahn]

(forgot to press the button) LGTM

[sanwou01]

I've adopted your suggestions, thanks!

[sanwou01]

Committed as https://reviews.llvm.org/rG93d9a4c95aff . Looks like I forgot to add the "Differential Revision:" line :(

llvm/lib/Passes/PassBuilder.cpp

 
 // Helper to add AnnotationRemarksPass.
 static void addAnnotationRemarksPass(ModulePassManager &MPM) {
   FunctionPassManager FPM;
   FPM.addPass(AnnotationRemarksPass());
   MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM)));
 }
 
+// Helper to check the current compilation phase
+static bool isLTOPreLink(ThinOrFullLTOPhase Phase) {
+  switch (Phase) {
+  case ThinOrFullLTOPhase::ThinLTOPreLink:
+  case ThinOrFullLTOPhase::FullLTOPreLink:
+    return true;
+  default:
+    return false;
+  }
+}
+
 // TODO: Investigate the cost/benefit of tail call elimination on debugging.
 FunctionPassManager
 PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level,
                                                    ThinOrFullLTOPhase Phase) {
 
   FunctionPassManager FPM(DebugLogging);
 
   // Form SSA out of local memory accesses after breaking apart aggregates into
PassBuilder::buildO1FunctionSimplificationPipeline(OptimizationLevel Level,
   LoopPassManager LPM1(DebugLogging), LPM2(DebugLogging);
 
   // Simplify the loop body. We do this initially to clean up after other loop
   // passes run, either when iterating on a loop or on inner loops with
   // implications on the outer loop.
   LPM1.addPass(LoopInstSimplifyPass());
   LPM1.addPass(LoopSimplifyCFGPass());
 
-  LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true));
+  LPM1.addPass(LoopRotatePass(/* Disable header duplication */ true,
+                              isLTOPreLink(Phase)));
   // TODO: Investigate promotion cap for O1.
   LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap));
   LPM1.addPass(SimpleLoopUnswitchPass());
 
   LPM2.addPass(LoopIdiomRecognizePass());
   LPM2.addPass(IndVarSimplifyPass());
 
   for (auto &C : LateLoopOptimizationsEPCallbacks)
PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level,
 
   // Simplify the loop body. We do this initially to clean up after other loop
   // passes run, either when iterating on a loop or on inner loops with
   // implications on the outer loop.
   LPM1.addPass(LoopInstSimplifyPass());
   LPM1.addPass(LoopSimplifyCFGPass());
 
   // Disable header duplication in loop rotation at -Oz.
-  LPM1.addPass(LoopRotatePass(Level != OptimizationLevel::Oz));
+  LPM1.addPass(
+      LoopRotatePass(Level != OptimizationLevel::Oz, isLTOPreLink(Phase)));
   // TODO: Investigate promotion cap for O1.
   LPM1.addPass(LICMPass(PTO.LicmMssaOptCap, PTO.LicmMssaNoAccForPromotionCap));
   LPM1.addPass(
       SimpleLoopUnswitchPass(/* NonTrivial */ Level == OptimizationLevel::O3 &&
                              EnableO3NonTrivialUnswitching));
   LPM2.addPass(LoopIdiomRecognizePass());
   LPM2.addPass(IndVarSimplifyPass());