[Pipeline] Don't run EarlyFPM in LTO post link

[aeubanks]

Summary

EarlyFPM cleans up the output of the frontend. This isn't necessary in post link pipelines as the pre link pipeline already ran this.

~0.4% savings in ThinLTO builds:
https://llvm-compile-time-tracker.com/compare.php?from=8a5d4eb775c644d8683f24817d44c510d2b853b7&to=3580252a2162eadca0da99f1eeaa112f74a0353d&stat=instructions:u

[aeubanks]

still need to update tests and make sure this doesn't regress anything, but sending out to make sure that this makes sense to you

[tejohnson]

In D145403#4172268, @aeubanks wrote:

still need to update tests and make sure this doesn't regress anything, but sending out to make sure that this makes sense to you

In theory this makes sense, certainly per the comment there. Did you see any surprising test changes?

[aeubanks]

update tests

[aeubanks]

In D145403#4175178, @tejohnson wrote:

In D145403#4172268, @aeubanks wrote:

still need to update tests and make sure this doesn't regress anything, but sending out to make sure that this makes sense to you

In theory this makes sense, certainly per the comment there. Did you see any surprising test changes?

no, just pipeline tests. I think this is good to go. I ran some benchmarks and didn't see any regressions, but I'll run them one more time before submitting

[tejohnson]

lgtm

llvm/lib/Passes/PassBuilderPipelines.cpp

985

I am reminded when reviewing this change that this method assumes it is not called during the FullLTOPostLink phase. Which it isn't today, but can you add an assert at the top of this function that the Phase is not FullLTOPostLink since all the checking here would break down if anything changed that?

[aeubanks]

assert not full lto postlink

[nikic]

Any reason why this hasn't been landed yet?

[aeubanks]

I was trying to get this landed with https://reviews.llvm.org/D145265 so I could measure the impact of the two together. But also now there are some merge conflicts that I'd rather deal with after submitting D145265. But that patch is more likely to get reverted anyway so yeah this should go in first.

[aeubanks]

rebase

llvm/lib/Passes/PassBuilderPipelines.cpp

PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level,
   // Ideally this should be determined from whether the IR is annotated with
   // sample profile, and not whether the a sample profile was provided on the
   // command line. E.g. for flattened profiles where we will not be reloading
   // the sample profile in the ThinLTO backend, we ideally shouldn't have to
   // provide the sample profile file.
   if (Phase == ThinOrFullLTOPhase::ThinLTOPostLink && !LoadSampleProfile)
     MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */, HasSampleProfile));
 
+  // Create an early function pass manager to cleanup the output of the
+  // frontend. Not necessary with LTO post link pipelines since the pre link
+  // pipeline already cleaned up the frontend output.
+  FunctionPassManager EarlyFPM;
+  if (Phase != ThinOrFullLTOPhase::ThinLTOPostLink) {
     // Do basic inference of function attributes from known properties of system
     // libraries and other oracles.
     MPM.addPass(InferFunctionAttrsPass());
     MPM.addPass(CoroEarlyPass());
 
-  // Create an early function pass manager to cleanup the output of the
-  // frontend.
-  FunctionPassManager EarlyFPM;
     // Lower llvm.expect to metadata before attempting transforms.
-  // Compare/branch metadata may alter the behavior of passes like SimplifyCFG.
+    // Compare/branch metadata may alter the behavior of passes like
+    // SimplifyCFG.
     EarlyFPM.addPass(LowerExpectIntrinsicPass());
     EarlyFPM.addPass(SimplifyCFGPass());
     EarlyFPM.addPass(SROAPass(SROAOptions::ModifyCFG));
     EarlyFPM.addPass(EarlyCSEPass());
     if (Level == OptimizationLevel::O3)
       EarlyFPM.addPass(CallSiteSplittingPass());
+  }
 
   // In SamplePGO ThinLTO backend, we need instcombine before profile annotation
   // to convert bitcast to direct calls so that they can be inlined during the
   // profile annotation prepration step.
   // More details about SamplePGO design can be found in:
   // https://research.google.com/pubs/pub45290.html
   // FIXME: revisit how SampleProfileLoad/Inliner/ICP is structured.
   if (LoadSampleProfile)
     EarlyFPM.addPass(InstCombinePass());
-  MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM),
-                                                PTO.EagerlyInvalidateAnalyses));
+
+  if (!EarlyFPM.isEmpty())
+    MPM.addPass(createModuleToFunctionPassAdaptor(
+        std::move(EarlyFPM), PTO.EagerlyInvalidateAnalyses));
 
   if (LoadSampleProfile) {
     // Annotate sample profile right after early FPM to ensure freshness of
     // the debug info.
     MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile,
                                         PGOOpt->ProfileRemappingFile, Phase));
     // Cache ProfileSummaryAnalysis once to avoid the potential need to insert
     // RequireAnalysisPass for PSI before subsequent non-module passes.