[SampleFDO] Support running sample loader in O0 mode

Author: wlei-llvm

llvm/lib/Passes/PassBuilderPipelines.cpp

@@ -302,6 +302,7 @@ static cl::opt<std::string> InstrumentColdFuncOnlyPath(
 
 extern cl::opt<std::string> UseCtxProfile;
 extern cl::opt<bool> PGOInstrumentColdFunctionOnly;
+extern cl::opt<bool> DisableSampleLoaderInlining;
 
 namespace llvm {
 extern cl::opt<bool> EnableMemProfContextDisambiguation;
@@ -2138,6 +2139,18 @@ ModulePassManager PassBuilder::buildO0DefaultPipeline(OptimizationLevel Level,
   if (PGOOpt && PGOOpt->DebugInfoForProfiling)
     MPM.addPass(createModuleToFunctionPassAdaptor(AddDiscriminatorsPass()));
 
+  if (PGOOpt && PGOOpt->Action == PGOOptions::SampleUse) {
+    // Explicitly disable sample loader inlining in O0 pipeline.
+    if (!DisableSampleLoaderInlining.getNumOccurrences())
+      DisableSampleLoaderInlining = true;
+    MPM.addPass(SampleProfileLoaderPass(PGOOpt->ProfileFile,
+                                        PGOOpt->ProfileRemappingFile,
+                                        ThinOrFullLTOPhase::None));
+    // Cache ProfileSummaryAnalysis once to avoid the potential need to insert
+    // RequireAnalysisPass for PSI before subsequent non-module passes.
+    MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>());
+  }
+
   invokePipelineEarlySimplificationEPCallbacks(MPM, Level);
 
   // Build a minimal pipeline based on the semantics required by LLVM,

llvm/lib/Transforms/IPO/SampleProfile.cpp

@@ -192,7 +192,7 @@ static cl::opt<bool> ProfileSizeInline(
 // Since profiles are consumed by many passes, turning on this option has
 // side effects. For instance, pre-link SCC inliner would see merged profiles
 // and inline the hot functions (that are skipped in this pass).
-static cl::opt<bool> DisableSampleLoaderInlining(
+cl::opt<bool> DisableSampleLoaderInlining(
     "disable-sample-loader-inlining", cl::Hidden, cl::init(false),
     cl::desc("If true, artifically skip inline transformation in sample-loader "
              "pass, and merge (or scale) profiles (as configured by "

llvm/test/Other/new-pm-pgo-O0.ll

@@ -9,15 +9,19 @@
 ; RUN:     |FileCheck %s --check-prefixes=USE_POST_LINK,USE
 ; RUN: opt -debug-pass-manager -passes='lto<O0>' -pgo-kind=pgo-instr-use-pipeline -profile-file='%t.profdata' %s 2>&1 \
 ; RUN:     |FileCheck %s --check-prefixes=USE_POST_LINK,USE
+; RUN: opt -debug-pass-manager -passes='default<O0>' -pgo-kind=pgo-sample-use-pipeline -profile-file='%S/Inputs/new-pm-pgo.prof' %s 2>&1 \
+; RUN:     |FileCheck %s --check-prefixes=SAMPLE_USE
 
-;
 ; GEN: Running pass: PGOInstrumentationGen
 ; USE_DEFAULT: Running pass: PGOInstrumentationUse
 ; USE_PRE_LINK: Running pass: PGOInstrumentationUse
 ; USE_POST_LINK-NOT: Running pass: PGOInstrumentationUse
 ; USE-NOT: Running pass: PGOIndirectCallPromotion
 ; USE-NOT: Running pass: PGOMemOPSizeOpt
 
+; SAMPLE_USE: Running pass: AddDiscriminatorsPass
+; SAMPLE_USE: Running pass: SampleProfileLoaderPass
+
 define void @foo() {
   ret void
 }