[PassBuilder] Don't use MemorySSA for standalone LoopRotate passes

[nikic]

Summary

Two standalone LoopRotate passes scheduled using createFunctionToLoopPassAdaptor() currently enable MemorySSA. However, while LoopRotate can preserve MemorySSA, it does not use it, so requiring MemorySSA is unnecessary.

Weirdly, I'm not seeing any significant compile-time change from this patch, so I'm wondering if I'm misunderstanding something here...

[nikic]

Fix clang pipeline test.

[nikic]

Weirdly, I'm not seeing any significant compile-time change from this patch, so I'm wondering if I'm misunderstanding something here...

The answer to that is D108074: There are subsequent passes that also have unnecessary MemorySSA requirements, so just dropping it here doesn't matter much in practice.

[asbirlea]

In D108073#2945148, @nikic wrote:

Weirdly, I'm not seeing any significant compile-time change from this patch, so I'm wondering if I'm misunderstanding something here...

The answer to that is D108074: There are subsequent passes that also have unnecessary MemorySSA requirements, so just dropping it here doesn't matter much in practice.

Makes sense. I'm happy to drop these. They can always be readded if those passes are changed to use (and preserve) MemorySSA.
Thank you for the cleanup patches!

clang/test/CodeGen/thinlto-distributed-newpm.ll

 ; CHECK-O: Running pass: ReversePostOrderFunctionAttrsPass
 ; CHECK-O: Running analysis: CallGraphAnalysis
 ; CHECK-O: Running pass: RequireAnalysisPass<{{.*}}GlobalsAA
 ; CHECK-O: Running pass: Float2IntPass on main
 ; CHECK-O: Running pass: LowerConstantIntrinsicsPass on main
 ; CHECK-O: Running pass: LoopSimplifyPass on main
 ; CHECK-O: Running analysis: LoopAnalysis on main
 ; CHECK-O: Running pass: LCSSAPass on main
-; CHECK-O: Running analysis: MemorySSAAnalysis on main
 ; CHECK-O: Running analysis: AAManager on main
 ; CHECK-O: Running analysis: BasicAA on main
 ; CHECK-O: Running analysis: ScalarEvolutionAnalysis on main
 ; CHECK-O: Running analysis: InnerAnalysisManagerProxy
 ; CHECK-O: Running pass: LoopRotatePass on Loop at depth 1 containing: %b
 ; CHECK-O: Running pass: LoopDistributePass on main
 ; CHECK-O: Running pass: InjectTLIMappings on main
 ; CHECK-O: Running pass: LoopVectorizePass on main
 ; CHECK-O: Running analysis: BlockFrequencyAnalysis on main
 ; CHECK-O: Running analysis: BranchProbabilityAnalysis on main
 ; CHECK-O: Running analysis: PostDominatorTreeAnalysis on main
 ; CHECK-O: Running analysis: DemandedBitsAnalysis on main
+; CHECK-O: Running analysis: MemorySSAAnalysis on main
 ; CHECK-O: Running pass: LoopLoadEliminationPass on main
 ; CHECK-O: Running pass: InstCombinePass on main
 ; CHECK-O: Running pass: SimplifyCFGPass on main
 ; CHECK-O: Running pass: SLPVectorizerPass on main
 ; CHECK-O: Running pass: VectorCombinePass on main
 ; CHECK-O: Running pass: InstCombinePass on main
 ; CHECK-O: Running pass: LoopUnrollPass on main
 ; CHECK-O: Running pass: WarnMissedTransformationsPass on main

llvm/lib/Passes/PassBuilder.cpp

void PassBuilder::addPGOInstrPasses(ModulePassManager &MPM,
   }
 
   // Perform PGO instrumentation.
   MPM.addPass(PGOInstrumentationGen(IsCS));
 
   FunctionPassManager FPM;
   // Disable header duplication in loop rotation at -Oz.
   FPM.addPass(createFunctionToLoopPassAdaptor(
-      LoopRotatePass(Level != OptimizationLevel::Oz), EnableMSSALoopDependency,
+      LoopRotatePass(Level != OptimizationLevel::Oz), /*UseMemorySSA=*/false,
       /*UseBlockFrequencyInfo=*/false));
   MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM)));
 
   // Add the profile lowering pass.
   InstrProfOptions Options;
   if (!ProfileFile.empty())
     Options.InstrProfileOutput = ProfileFile;
   // Do counter promotion at Level greater than O0.
PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level,
 
   for (auto &C : VectorizerStartEPCallbacks)
     C(OptimizePM, Level);
 
   // First rotate loops that may have been un-rotated by prior passes.
   // Disable header duplication at -Oz.
   OptimizePM.addPass(createFunctionToLoopPassAdaptor(
       LoopRotatePass(Level != OptimizationLevel::Oz, LTOPreLink),
-      EnableMSSALoopDependency,
-      /*UseBlockFrequencyInfo=*/false));
+      /*UseMemorySSA=*/false, /*UseBlockFrequencyInfo=*/false));
 
   // Distribute loops to allow partial vectorization.  I.e. isolate dependences
   // into separate loop that would otherwise inhibit vectorization.  This is
   // currently only performed for loops marked with the metadata
   // llvm.loop.distribute=true or when -enable-loop-distribute is specified.
   OptimizePM.addPass(LoopDistributePass());
 
   // Populates the VFABI attribute with the scalar-to-vector mappings