[llvm-branch-commits] [llvm] ff7e1da - [NPM] Support -fmerge-functions

[Arthur Eubanks via llvm-branch-commits]

Author: Arthur Eubanks
Date: 2020-12-10T11:45:08-08:00
New Revision: ff7e1da68f2a74797d0d3454f6cac62064f7d982

URL: https://github.com/llvm/llvm-project/commit/ff7e1da68f2a74797d0d3454f6cac62064f7d982
DIFF: https://github.com/llvm/llvm-project/commit/ff7e1da68f2a74797d0d3454f6cac62064f7d982.diff

LOG: [NPM] Support -fmerge-functions

I tried to put it in the same place in the pipeline as the legacy PM.

Fixes PR48399.

Reviewed By: asbirlea, nikic

Differential Revision: https://reviews.llvm.org/D93002

Added: 
    

Modified: 
    clang/lib/CodeGen/BackendUtil.cpp
    clang/test/CodeGenCXX/merge-functions.cpp
    llvm/include/llvm/Passes/PassBuilder.h
    llvm/lib/Passes/PassBuilder.cpp

Removed: 
    


################################################################################
diff  --git a/clang/lib/CodeGen/BackendUtil.cpp b/clang/lib/CodeGen/BackendUtil.cpp
index bde4c25c7bc2..554688ac987b 100644
--- a/clang/lib/CodeGen/BackendUtil.cpp
+++ b/clang/lib/CodeGen/BackendUtil.cpp
@@ -1139,6 +1139,7 @@ void EmitAssemblyHelper::EmitAssemblyWithNewPassManager(
   PTO.LoopInterleaving = CodeGenOpts.UnrollLoops;
   PTO.LoopVectorization = CodeGenOpts.VectorizeLoop;
   PTO.SLPVectorization = CodeGenOpts.VectorizeSLP;
+  PTO.MergeFunctions = CodeGenOpts.MergeFunctions;
   // Only enable CGProfilePass when using integrated assembler, since
   // non-integrated assemblers don't recognize .cgprofile section.
   PTO.CallGraphProfile = !CodeGenOpts.DisableIntegratedAS;

diff  --git a/clang/test/CodeGenCXX/merge-functions.cpp b/clang/test/CodeGenCXX/merge-functions.cpp
index db742f41d3c8..42d57d7db811 100644
--- a/clang/test/CodeGenCXX/merge-functions.cpp
+++ b/clang/test/CodeGenCXX/merge-functions.cpp
@@ -1,5 +1,8 @@
 // REQUIRES: x86-registered-target
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -O0 -fno-experimental-new-pass-manager -fmerge-functions -emit-llvm -o - -x c++ < %s | FileCheck %s -implicit-check-not=_ZN1A1gEiPi
 // RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -O1 -fno-experimental-new-pass-manager -fmerge-functions -emit-llvm -o - -x c++ < %s | FileCheck %s -implicit-check-not=_ZN1A1gEiPi
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -O0 -fexperimental-new-pass-manager -fmerge-functions -emit-llvm -o - -x c++ < %s | FileCheck %s -implicit-check-not=_ZN1A1gEiPi
+// RUN: %clang_cc1 -triple x86_64-pc-linux-gnu -O1 -fexperimental-new-pass-manager -fmerge-functions -emit-llvm -o - -x c++ < %s | FileCheck %s -implicit-check-not=_ZN1A1gEiPi
 
 // Basic functionality test. Function merging doesn't kick in on functions that
 // are too simple.

diff  --git a/llvm/include/llvm/Passes/PassBuilder.h b/llvm/include/llvm/Passes/PassBuilder.h
index a04e98df8f85..e2d22031dd5e 100644
--- a/llvm/include/llvm/Passes/PassBuilder.h
+++ b/llvm/include/llvm/Passes/PassBuilder.h
@@ -123,6 +123,10 @@ class PipelineTuningOptions {
   /// Tuning option to enable/disable call graph profile. Its default value is
   /// that of the flag: `-enable-npm-call-graph-profile`.
   bool CallGraphProfile;
+
+  /// Tuning option to enable/disable function merging. Its default value is
+  /// false.
+  bool MergeFunctions;
 };
 
 /// This class provides access to building LLVM's passes.

diff  --git a/llvm/lib/Passes/PassBuilder.cpp b/llvm/lib/Passes/PassBuilder.cpp
index 0e29b14dac9a..bb77124e48d6 100644
--- a/llvm/lib/Passes/PassBuilder.cpp
+++ b/llvm/lib/Passes/PassBuilder.cpp
@@ -282,6 +282,7 @@ PipelineTuningOptions::PipelineTuningOptions() {
   LicmMssaOptCap = SetLicmMssaOptCap;
   LicmMssaNoAccForPromotionCap = SetLicmMssaNoAccForPromotionCap;
   CallGraphProfile = true;
+  MergeFunctions = false;
 }
 
 extern cl::opt<bool> EnableConstraintElimination;
@@ -1316,6 +1317,10 @@ PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level,
   if (EnableHotColdSplit && !LTOPreLink)
     MPM.addPass(HotColdSplittingPass());
 
+  // Merge functions if requested.
+  if (PTO.MergeFunctions)
+    MPM.addPass(MergeFunctionsPass());
+
   // LoopSink pass sinks instructions hoisted by LICM, which serves as a
   // canonicalization pass that enables other optimizations. As a result,
   // LoopSink pass needs to be a very late IR pass to avoid undoing LICM
@@ -1746,10 +1751,12 @@ PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level,
   // Now that we have optimized the program, discard unreachable functions.
   MPM.addPass(GlobalDCEPass());
 
+  if (PTO.MergeFunctions)
+    MPM.addPass(MergeFunctionsPass());
+
   // Emit annotation remarks.
   addAnnotationRemarksPass(MPM);
 
-  // FIXME: Maybe enable MergeFuncs conditionally after it's ported.
   return MPM;
 }
 
@@ -1781,6 +1788,9 @@ ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level,
   MPM.addPass(AlwaysInlinerPass(
       /*InsertLifetimeIntrinsics=*/PTO.Coroutines));
 
+  if (PTO.MergeFunctions)
+    MPM.addPass(MergeFunctionsPass());
+
   if (EnableMatrix)
     MPM.addPass(
         createModuleToFunctionPassAdaptor(LowerMatrixIntrinsicsPass(true)));