Index: llvm/trunk/lib/Transforms/Instrumentation/PGOInstrumentation.cpp =================================================================== --- llvm/trunk/lib/Transforms/Instrumentation/PGOInstrumentation.cpp +++ llvm/trunk/lib/Transforms/Instrumentation/PGOInstrumentation.cpp @@ -182,6 +182,14 @@ cl::desc("Use this option to turn on/off " "memory intrinsic size profiling.")); +// Emit branch probability as optimization remarks. +static cl::opt + EmitBranchProbability("pgo-emit-branch-prob", cl::init(false), cl::Hidden, + cl::desc("When this option is on, the annotated " + "branch probability will be emitted as " + " optimization remarks: -Rpass-analysis=" + "pgo-instr-use")); + // Command line option to turn on CFG dot dump after profile annotation. // Defined in Analysis/BlockFrequencyInfo.cpp: -pgo-view-counts extern cl::opt PGOViewCounts; @@ -192,6 +200,39 @@ namespace { +// Return a string describing the branch condition that can be +// used in static branch probability heuristics: +std::string getBranchCondString(Instruction *TI) { + BranchInst *BI = dyn_cast(TI); + if (!BI || !BI->isConditional()) + return std::string(); + + Value *Cond = BI->getCondition(); + ICmpInst *CI = dyn_cast(Cond); + if (!CI) + return std::string(); + + std::string result; + raw_string_ostream OS(result); + OS << CmpInst::getPredicateName(CI->getPredicate()) << "_"; + CI->getOperand(0)->getType()->print(OS, true); + + Value *RHS = CI->getOperand(1); + ConstantInt *CV = dyn_cast(RHS); + if (CV) { + if (CV->isZero()) + OS << "_Zero"; + else if (CV->isOne()) + OS << "_One"; + else if (CV->isAllOnesValue()) + OS << "_MinusOne"; + else + OS << "_Const"; + } + OS.flush(); + return result; +} + /// The select instruction visitor plays three roles specified /// by the mode. In \c VM_counting mode, it simply counts the number of /// select instructions. In \c VM_instrument mode, it inserts code to count @@ -1424,6 +1465,29 @@ for (const auto &W : Weights) { dbgs() << W << " "; } dbgs() << "\n";); TI->setMetadata(llvm::LLVMContext::MD_prof, MDB.createBranchWeights(Weights)); + if (EmitBranchProbability) { + std::string BrCondStr = getBranchCondString(TI); + if (BrCondStr.empty()) + return; + + unsigned WSum = + std::accumulate(Weights.begin(), Weights.end(), 0, + [](unsigned w1, unsigned w2) { return w1 + w2; }); + uint64_t TotalCount = + std::accumulate(EdgeCounts.begin(), EdgeCounts.end(), 0, + [](uint64_t c1, uint64_t c2) { return c1 + c2; }); + BranchProbability BP(Weights[0], WSum); + std::string BranchProbStr; + raw_string_ostream OS(BranchProbStr); + OS << BP; + OS << " (total count : " << TotalCount << ")"; + OS.flush(); + Function *F = TI->getParent()->getParent(); + emitOptimizationRemarkAnalysis( + F->getContext(), "pgo-use-annot", *F, TI->getDebugLoc(), + Twine(BrCondStr) + + " is true with probability : " + Twine(BranchProbStr)); + } } template <> struct GraphTraits { Index: llvm/trunk/test/Transforms/PGOProfile/branch1.ll =================================================================== --- llvm/trunk/test/Transforms/PGOProfile/branch1.ll +++ llvm/trunk/test/Transforms/PGOProfile/branch1.ll @@ -15,6 +15,9 @@ ; RUN: opt < %s -passes=pgo-instr-use -pgo-test-profile-file=%t.profdata -S | FileCheck %s --check-prefix=USE ; RUN: opt < %s -passes=pgo-instr-use -pgo-test-profile-file=%t.l.profdata -S | FileCheck %s --check-prefix=USE-LARGE +; RUN: opt < %s -pgo-instr-use -pgo-test-profile-file=%t.profdata -pass-remarks-analysis=pgo-use-annot -pgo-emit-branch-prob -S 2>&1| FileCheck %s --check-prefix=ANALYSIS +; RUN: opt < %s -passes=pgo-instr-use -pgo-test-profile-file=%t.profdata -pass-remarks-analysis=pgo-use-annot -pgo-emit-branch-prob -S 2>&1| FileCheck %s --check-prefix=ANALYSIS + target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" target triple = "x86_64-unknown-linux-gnu" ; GEN-DARWIN-LINKONCE: target triple = "x86_64-apple-darwin" @@ -54,3 +57,5 @@ ; USE-DAG: {{![0-9]+}} = !{i32 1, !"ProfileSummary", {{![0-9]+}}} ; USE-DAG: {{![0-9]+}} = !{!"DetailedSummary", {{![0-9]+}}} ; USE-DAG: ![[FUNC_ENTRY_COUNT]] = !{!"function_entry_count", i64 3} + +; ANALYSIS:remark: :0:0: sgt_i32_Zero {{.*}}66.67% (total count : 3)