[PM] Flesh out the new pm LTO pipeline

[davide]

Summary

This is pretty much a carbon copy of the current pass manager pipeline, and it unblocks my testing. Passes not yet ported or disabled because of problems with the loop PM are marked with a FIXME. So, not complete, but hopefully a starting point. For those watching, this is so that I can start playing with PGO support in the new inliner (at LTO time).

[davide]

Note: this needs some tests associated, but before spending time on that I want to make sure we agree with the general direction.

[mehdi_amini]

What is the justification instead of adopting the ThinLTO pipeline?

[davide]

In D28996#652797, @mehdi_amini wrote:

What is the justification instead of adopting the ThinLTO pipeline?

The two problems are largely orthogonal. This is a port of the old LTO pipeline to the new pass manager. The aim is to make sure the current LTO pipeline works correctly with the new pass manager. I prefer to keep the old functionality in the new pass manager until it stabilizes. After that (or in parallel), we can decide whether we want to unify the two pipelines (if it's possible/reasonable).
On top of that, currently plugging the ThinLTO pipeline for LTO results in 2x/2.5x increase in compile time. Definitely not an hit I want to pay.

[mehdi_amini]

In D28996#652801, @davide wrote:

In D28996#652797, @mehdi_amini wrote:

What is the justification instead of adopting the ThinLTO pipeline?

The two problems are largely orthogonal. This is a port of the old LTO pipeline to the new pass manager.

Apologies, I totally misunderstood what you were doing here, I thought this was the followup to our discussion last week about revisiting the LTO PM.

[mehdi_amini]

LGTM.
(Not sure about testing, may not be worth it at this point).

lib/Passes/PassBuilder.cpp

553

Alias analyses are added one layer above IIRC?

575

This seems dead?

619

It seems that there is "PruneEH" in the legacy PM here?

[davide]

lib/Passes/PassBuilder.cpp

553

If you mean, in the caller of buildLTODefaultPipeline, then answer is yes (see my other patch for lib/LTO.

575

Yes, nice catch. I originally created this one as I thought we were running multiple CGSCC passes, but it turns out we just run PostOrderFunctionAttrs so that can be folded into the only argument of createModuleToPostOrderCGSCCPassAdaptor().

619

We used to. But it wasn't ported to the new PM and Chandler is not entirely convinced it's worth running this anymore (see the comment in the perModulePipeline).
Happy to duplicate the comment here, or add a FIXME.
Side note: I have no strong opinion whether we should run the pass or not.

[chandlerc]

I would suggest basic testing here similar to what we do for O[1-3sz] in test/Other/new-pm-defaults.ll. I'd keep the pre-link-LTO and non-LTO in that file and add a new file (new-pm-lto-defaults.ll) for the post-link LTO pipeline.

lib/Passes/PassBuilder.cpp

672

Typo here: assumptions -> Assumptions

[davide]

Added tests as per Chandler's suggestion and addressed Mehdi's comments.

[chandlerc]

test/Other/new-pm-lto-defaults.ll

22 ↗

(On Diff #85421)

Why not use -NEXT checks the way mkuper made the main one? Seems much more robust at detecting changes and making sure they are intentional.

[davide]

Use CHECK-NEXT for more robust testing as recommended by Chandler.

[chandlerc]

LGTM, thanks!

lib/Passes/PassBuilder.cpp

   return buildPerModuleDefaultPipeline(Level, DebugLogging);
 }
 
 ModulePassManager PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level,
                                                        bool DebugLogging) {
   assert(Level != O0 && "Must request optimizations for the default pipeline!");
   ModulePassManager MPM(DebugLogging);
 
-  // FIXME: Finish fleshing this out to match the legacy LTO pipelines.
-  FunctionPassManager LateFPM(DebugLogging);
-  LateFPM.addPass(InstCombinePass());
-  LateFPM.addPass(SimplifyCFGPass());
+  // Remove unused virtual tables to improve the quality of code generated by
+  // whole-program devirtualization and bitset lowering.
+  MPM.addPass(GlobalDCEPass());
+
+  // Force any function attributes we want the rest of the pipeline to observe.
+  MPM.addPass(ForceFunctionAttrsPass());
+
+  // Do basic inference of function attributes from known properties of system
+  // libraries and other oracles.
+  MPM.addPass(InferFunctionAttrsPass());
+
+  if (Level > 1) {
+    // Indirect call promotion. This should promote all the targets that are
+    // left by the earlier promotion pass that promotes intra-module targets.
+    // This two-step promotion is to save the compile time. For LTO, it should
+    // produce the same result as if we only do promotion here.
+    MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */));
+
+    // Propagate constants at call sites into the functions they call.  This
+    // opens opportunities for globalopt (and inlining) by substituting function
+    // pointers passed as arguments to direct uses of functions.
+    MPM.addPass(IPSCCPPass());
+  }
+
+  // Small CGSCC pipeline to run PostOrderFunctionAttr pass.
+  CGSCCPassManager PostOrderCGPipeline(DebugLogging);
+
+  // Now deduce any function attributes based in the current code.
+  MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(
+              PostOrderFunctionAttrsPass()));
+
+  // Do RPO function attribute inference across the module to forward-propagate
+  // attributes where applicable.
+  // FIXME: Is this really an optimization rather than a canonicalization?
+  MPM.addPass(ReversePostOrderFunctionAttrsPass());
+
+  // Use inragne annotations on GEP indices to split globals where beneficial.
+  MPM.addPass(GlobalSplitPass());
+
+  // Run whole program optimization of virtual call when the list of callees
+  // is fixed.
+  MPM.addPass(WholeProgramDevirtPass());
+
+  // Stop here at -O1.
+  if (Level == 1)
+    return MPM;
+
+  // Optimize globals to try and fold them into constants.
+  MPM.addPass(GlobalOptPass());
+
+  // Promote any localized globals to SSA registers.
+  MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass()));
+
+  // Linking modules together can lead to duplicate global constant, only
+  // keep one copy of each constant.
+  MPM.addPass(ConstantMergePass());
+
+  // Remove unused arguments from functions.
+  MPM.addPass(DeadArgumentEliminationPass());
+
+  // Reduce the code after globalopt and ipsccp.  Both can open up significant
+  // simplification opportunities, and both can propagate functions through
+  // function pointers.  When this happens, we often have to resolve varargs
+  // calls, etc, so let instcombine do this.
+  // FIXME: add peephole extensions here as the legacy PM does.
+  MPM.addPass(createModuleToFunctionPassAdaptor(InstCombinePass()));
+
+  // Run the inliner now.
+  MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(InlinerPass()));
 
-  MPM.addPass(createModuleToFunctionPassAdaptor(std::move(LateFPM)));
+  // Optimize globals again after we ran the inliner.
+  MPM.addPass(GlobalOptPass());
+
+  // Garbage collect dead functions.
+  // FIXME: Add ArgumentPromotion pass after once it's ported.
+  MPM.addPass(GlobalDCEPass());
+
+  FunctionPassManager FPM(DebugLogging);
+
+  // The IPO Passes may leave cruft around. Clean up after them.
+  // FIXME: add peephole extensions here as the legacy PM does.
+  FPM.addPass(InstCombinePass());
+  FPM.addPass(JumpThreadingPass());
+
+  // Break up allocas
+  FPM.addPass(SROA());
+
+  // Run a few AA driver optimizations here and now to cleanup the code.
+  MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM)));
+
+  MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(
+              PostOrderFunctionAttrsPass()));
+  // FIXME: here we run IP alias analysis in the legacy PM.
+
+  FunctionPassManager MainFPM;
+
+  // FIXME: once we fix LoopPass Manager, add LICM here.
+  // FIXME: once we provide support for enabling MLSM, add it here.
+  // FIXME: once we provide support for enabling NewGVN, add it here.
+  MainFPM.addPass(GVN());
+
+  // Remove dead memcpy()'s.
+  MainFPM.addPass(MemCpyOptPass());
+
+  // Nuke dead stores.
+  MainFPM.addPass(DSEPass());
+
+  // FIXME: at this point, we run a bunch of loop passes:
+  // indVarSimplify, loopDeletion, loopInterchange, loopUnrool,
+  // loopVectorize. Enable them once the remaining issue with LPM
+  // are sorted out.
+
+  MainFPM.addPass(InstCombinePass());
+  MainFPM.addPass(SimplifyCFGPass());
+  MainFPM.addPass(SCCPPass());
+  MainFPM.addPass(InstCombinePass());
+  MainFPM.addPass(BDCEPass());
+
+  // FIXME: We may want to run SLPVectorizer here.
+  // After vectorization, assume intrinsics may tell us more
+  // about pointer alignments.
+#if 0
+  MainFPM.add(AlignmentFromassumptionsPass());
+#endif
+
+  // FIXME: Conditionally run LoadCombine here, after it's ported
+  // (in case we still have this pass, given its questionable usefulness).
+
+  // FIXME: add peephole extensions to the PM here.
+  MainFPM.addPass(InstCombinePass());
+  MainFPM.addPass(JumpThreadingPass());
+  MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM)));
+
+  // Create a function that performs CFI checks for cross-DSO calls with
+  // targets in the current module.
+  MPM.addPass(CrossDSOCFIPass());
+
+  // Lower type metadata and the type.test intrinsic. This pass supports
+  // clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs
+  // to be run at link time if CFI is enabled. This pass does nothing if
+  // CFI is disabled.
+  // Enable once we add support for the summary in the new PM.
+#if 0
+  MPM.addPass(LowerTypeTestsPass(Summary ? LowerTypeTestsSummaryAction::Export :
+                                           LowerTypeTestsSummaryAction::None,
+                                Summary));
+#endif
+
+  // Add late LTO optimization passes.
+  // Delete basic blocks, which optimization passes may have killed.
+  MPM.addPass(createModuleToFunctionPassAdaptor(SimplifyCFGPass()));
+
+  // Drop bodies of available eternally objects to improve GlobalDCE.
+  MPM.addPass(EliminateAvailableExternallyPass());
+
+  // Now that we have optimized the program, discard unreachable functions.
+  MPM.addPass(GlobalDCEPass());
 
+  // FIXME: Enable MergeFuncs, conditionally, after ported, maybe.
   return MPM;
 }
 
 AAManager PassBuilder::buildDefaultAAPipeline() {
   AAManager AA;
 
   // The order in which these are registered determines their priority when
   // being queried.