[SelectOpti][4/5] Loop Heuristics

[apostolakis]

Summary

This patch adds the loop-level heuristics for determining whether branches are more profitable than conditional moves.
These heuristics apply to only inner-most loops.

Depends on D120231

[apostolakis]

Merge

[apostolakis]

Rebase and move tests

[apostolakis]

Use opt tests instead of llc ones

[apostolakis]

Rebase

[apostolakis]

Rebase and add more comments for methods

[davidxl]

A high level comment. I saw ScalingUpFactor is used in many places. It may be better to directly use Scaled64 data type instead of uint64 to avoid that.

[apostolakis]

Use Scaled64 instead of using ScalingUpFactor.

[apostolakis]

In D120232#3460939, @davidxl wrote:

A high level comment. I saw ScalingUpFactor is used in many places. It may be better to directly use Scaled64 data type instead of uint64 to avoid that.

I used ScalingUpFactor to get a decent precision without increasing compile time (due to non-integer ops), but it is a bit hacky and it seems better to directly use Scaled64.

[davidxl]

llvm/lib/CodeGen/SelectOptimize.cpp

464

Not clear about what the second condition actually means. What is the cost of select group?

474

Should this be missed optimization or Analysis. The message probably also needs more context to be useful.

667

Add a parameter?

672

Add explanation of this heuristic.

682

Document the return value.

685

Perhaps passing by pointer explicitly?

689

why only 2 iterations?

717

Should this exclude the selection instruction itself for non-pred case?

745

should it weighted with misprediction rate?

749

Should Sum (of cost in a SI group) be used instead of max?

791

Should it be scaled by CondCost instead of taking the max?

811

Should there be an early return here?

[apostolakis]

llvm/lib/CodeGen/SelectOptimize.cpp

464

Yes that was unclear. Clarified the second condition.

474

I would prefer to stick to miss and pass remarks and not add analysis. It seems to me more natural and easier to know what to query for.
Changed though the code to emit more fine-grained and informative remarks for these cases.

667

Added.

672

Added.

682

Done.

685

Sure. Changed it.

689

Two iterations suffice to determine if the the loop's critical path involves loop-carried dependences and compute the rate that the critical path increases from one iteration to another.

717

It would be correct to exclude the select instructions from here but for the select instructions the INonPredCost is overwritten later on. So, we can skip adding a conditional here.

745

Actually it is weighted with the misprediction rate:
MispredictCost = max(MispredictPenalty, CondCost) * MispredictRate

749

This computes the critical path which essentially is the cost of the last instruction of the critical path (and thus the instruction with the highest latency). This is regardless of how many select groups are included in the loop. These groups may or may not be part of the path.

791

I do not think it should be scaled. The cost we want to compute are the cycles lost due to flushing the pipeline after a mispredict. If the CondCost is small, the operands of the condition are ready when needed and the MispredictPenalty (which depends on the pipeline depth and where the branch condition is determined in the pipeline) captures the mispredict cost . If the CondCost is very big, then the cycles lost end up being the latency of determining the condition, which is essentially CondCost. For in-between cases, the correct answer is not really the max but there is still some overlap. Therefore, max seemed a reasonable choice and it also seemed to be experimentally the best option among other versions I tested.

811

Good catch! Yes, erased by mistake the early return when I converted to Scaled64.

[apostolakis]

Address review comments.

[davidxl]

llvm/lib/CodeGen/SelectOptimize.cpp

466

The second heuristic is similar to the first one -- they are related to latency. Does it make more sense to consider the total pressure on resources?

Also it will be helpful to document in which function herusitc i and ii) are implemented.

673

Perhaps name Diff to Gain to indicate it is the gain from converting Pred to NonPred.

llvm/test/CodeGen/X86/select-optimize.ll

215

Add more test cases to cover other heuristics like gradiant

[apostolakis]

llvm/lib/CodeGen/SelectOptimize.cpp

466

Yes both heuristics are related to latency. But the first one is at a loop-level and all select groups are considered together (if all remain selects or all converted to branches; would be too expensive to try all different combinations), while the second one examines each select group separately. There are cases where one particular select group is unprofitable (or unnecessary) to convert while all the other ones should be converted. The second check allows us to prevent a conversion on the unprofitable one or save some code size if there is no perf diff.

I think considering total pressure on resources would be a separate patch that needs some exploration.

Improved the comments to clarify and added some pointers as suggested.

673

Good point. Changed it.
I actually also added more info in the opti remarks for this part of the code to make tuning of all the thresholds easier.

llvm/test/CodeGen/X86/select-optimize.ll

215

Yep I should have more testing coverage. Added.

[apostolakis]

Address comments with more testing and tweaks.

[bsmith]

llvm/lib/CodeGen/SelectOptimize.cpp

749

Is this really what we want though? As an example, a workload I'm looking at has a structure akin to:

----------------------------critical path
     \------------select-store
     \------------select-store
     \------------select-store
     \------------select-store
     \------------select-store

None of the selects are on the critical path, however they are still significant in terms of cost. This may be for a future patch that looks at resources, but it may be better to cost each path with a select on it individually, rather than costing the loop as a whole.

[apostolakis]

llvm/lib/CodeGen/SelectOptimize.cpp

749

There are definitely more cases that we might want to consider beyond the critical path. Yet, focusing on the critical path for this initial series of this new optimization will help catch the most obvious patterns and reduce the chances of regressions. Cases such as the one you mention will indeed need modeling of the resources and more testing, since there are definitely cases where you want to avoid a branch for non-critical-path selects.

I could add an option to disable the loop-level heuristics (that checks the reduction in the critical path, see line 462) and thus reduce the decision to the cost of each individual select (see line 465 and 478). This will allow experimenting with some of these cases. These costs will capture the gains for all the paths but it will not explicitly capture the total cost of each path. For your case, it will report the gain of converting the selects to branches (new vs old select cost) but it will not report the total cost of each path, i.e., select_cost + store_cost. Could try to add monitoring of all the local maximum in the loop and report them if you want.

[bsmith]

llvm/lib/CodeGen/SelectOptimize.cpp

749

An option to disable the critical path check would be good in the meantime I think. If I manually remove that check then this patch series, alongside the other patch you have to sink instructions, provide a very significant speedup for my case.

[apostolakis]

llvm/lib/CodeGen/SelectOptimize.cpp

749

Added the option to disable the critical path check.

[apostolakis]

Add option to disable the critical path check.

[davidxl]

lgtm

llvm/lib/CodeGen/SelectOptimize.cpp

 //===--- SelectOptimize.cpp - Convert select to branches if profitable ---===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This pass converts selects to conditional jumps when profitable.
 //
 //===----------------------------------------------------------------------===//
 
+#include "llvm/ADT/Optional.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/Analysis/BlockFrequencyInfo.h"
 #include "llvm/Analysis/BranchProbabilityInfo.h"
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/OptimizationRemarkEmitter.h"
 #include "llvm/Analysis/ProfileSummaryInfo.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/TargetLowering.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/CodeGen/TargetSchedule.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/Instruction.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Pass.h"
+#include "llvm/Support/ScaledNumber.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Transforms/Utils/SizeOpts.h"
 #include <algorithm>
 #include <memory>
 #include <queue>
 #include <stack>
 #include <string>
 
 using namespace llvm;
 
 #define DEBUG_TYPE "select-optimize"
 
 STATISTIC(NumSelectOptAnalyzed,
           "Number of select groups considered for conversion to branch");
 STATISTIC(NumSelectConvertedExpColdOperand,
           "Number of select groups converted due to expensive cold operand");
 STATISTIC(NumSelectConvertedHighPred,
           "Number of select groups converted due to high-predictability");
 STATISTIC(NumSelectUnPred,
           "Number of select groups not converted due to unpredictability");
 STATISTIC(NumSelectColdBB,
           "Number of select groups not converted due to cold basic block");
+STATISTIC(NumSelectConvertedLoop,
+          "Number of select groups converted due to loop-level analysis");
 STATISTIC(NumSelectsConverted, "Number of selects converted");
 
 static cl::opt<unsigned> ColdOperandThreshold(
     "cold-operand-threshold",
     cl::desc("Maximum frequency of path for an operand to be considered cold."),
     cl::init(20), cl::Hidden);
 
 static cl::opt<unsigned> ColdOperandMaxCostMultiplier(
     "cold-operand-max-cost-multiplier",
     cl::desc("Maximum cost multiplier of TCC_expensive for the dependence "
              "slice of a cold operand to be considered inexpensive."),
     cl::init(1), cl::Hidden);
 
+static cl::opt<unsigned>
+    GainGradientThreshold("select-opti-loop-gradient-gain-threshold",
+                          cl::desc("Gradient gain threshold (%)."),
+                          cl::init(25), cl::Hidden);
+
+static cl::opt<unsigned>
+    GainCycleThreshold("select-opti-loop-cycle-gain-threshold",
+                       cl::desc("Minimum gain per loop (in cycles) threshold."),
+                       cl::init(4), cl::Hidden);
+
+static cl::opt<unsigned> GainRelativeThreshold(
+    "select-opti-loop-relative-gain-threshold",
+    cl::desc(
+        "Minimum relative gain per loop threshold (1/X). Defaults to 12.5%"),
+    cl::init(8), cl::Hidden);
+
+static cl::opt<unsigned> MispredictDefaultRate(
+    "mispredict-default-rate", cl::Hidden, cl::init(25),
+    cl::desc("Default mispredict rate (initialized to 25%)."));
+
 namespace {
 
 class SelectOptimize : public FunctionPass {
   const TargetMachine *TM = nullptr;
   const TargetSubtargetInfo *TSI;
   const TargetLowering *TLI = nullptr;
   const TargetTransformInfo *TTI = nullptr;
   const LoopInfo *LI;
   DominatorTree *DT;
   std::unique_ptr<BlockFrequencyInfo> BFI;
   std::unique_ptr<BranchProbabilityInfo> BPI;
   ProfileSummaryInfo *PSI;
   OptimizationRemarkEmitter *ORE;
+  TargetSchedModel TSchedModel;
 
 public:
   static char ID;
 
   SelectOptimize() : FunctionPass(ID) {
     initializeSelectOptimizePass(*PassRegistry::getPassRegistry());
   }
 
public:
   }
 
 private:
   // Select groups consist of consecutive select instructions with the same
   // condition.
   using SelectGroup = SmallVector<SelectInst *, 2>;
   using SelectGroups = SmallVector<SelectGroup, 2>;
 
+  using Scaled64 = ScaledNumber<uint64_t>;
+
+  struct CostInfo {
+    /// Predicated cost (with selects as conditional moves).
+    Scaled64 PredCost;
+    /// Non-predicated cost (with selects converted to branches).
+    Scaled64 NonPredCost;
+  };
+
   // Converts select instructions of a function to conditional jumps when deemed
   // profitable. Returns true if at least one select was converted.
   bool optimizeSelects(Function &F);
 
   // Heuristics for determining which select instructions can be profitably
   // conveted to branches. Separate heuristics for selects in inner-most loops
   // and the rest of code regions (base heuristics for non-inner-most loop
   // regions).
   void optimizeSelectsBase(Function &F, SelectGroups &ProfSIGroups);
   void optimizeSelectsInnerLoops(Function &F, SelectGroups &ProfSIGroups);
 
   // Converts to branches the select groups that were deemed
   // profitable-to-convert.
   void convertProfitableSIGroups(SelectGroups &ProfSIGroups);
 
   // Splits selects of a given basic block into select groups.
   void collectSelectGroups(BasicBlock &BB, SelectGroups &SIGroups);
 
   // Determines for which select groups it is profitable converting to branches
-  // (base heuristics).
+  // (base and inner-most-loop heuristics).
   void findProfitableSIGroupsBase(SelectGroups &SIGroups,
                                   SelectGroups &ProfSIGroups);
+  void findProfitableSIGroupsInnerLoops(const Loop *L, SelectGroups &SIGroups,
+                                        SelectGroups &ProfSIGroups);
+
   // Determines if a select group should be converted to a branch (base
   // heuristics).
   bool isConvertToBranchProfitableBase(const SmallVector<SelectInst *, 2> &ASI);
 
   // Returns true if there are expensive instructions in the cold value
   // operand's (if any) dependence slice of any of the selects of the given
   // group.
   bool hasExpensiveColdOperand(const SmallVector<SelectInst *, 2> &ASI);
 
   // For a given source instruction, collect its backwards dependence slice
   // consisting of instructions exclusively computed for producing the operands
   // of the source instruction.
   void getExclBackwardsSlice(Instruction *I,
                              SmallVector<Instruction *, 2> &Slice);
 
   // Returns true if the condition of the select is highly predictable.
   bool isSelectHighlyPredictable(const SelectInst *SI);
 
+  // Loop-level checks to determine if a non-predicated version (with branches)
+  // of the given loop is more profitable than its predicated version.
+  bool checkLoopHeuristics(const Loop *L, const CostInfo LoopDepth[2]);
+
+  // Computes instruction and loop-critical-path costs for both the predicated
+  // and non-predicated version of the given loop.
+  bool computeLoopCosts(const Loop *L, const SelectGroups &SIGroups,
+                        DenseMap<const Instruction *, CostInfo> &InstCostMap,
+                        CostInfo *LoopCost);
+
+  // Returns a set of all the select instructions in the given select groups.
+  SmallPtrSet<const Instruction *, 2> getSIset(const SelectGroups &SIGroups);
+
+  // Returns the latency cost of a given instruction.
+  Optional<uint64_t> computeInstCost(const Instruction *I);
+
+  // Returns the misprediction cost of a given select when converted to branch.
+  Scaled64 getMispredictionCost(const SelectInst *SI, const Scaled64 CondCost);
+
+  // Returns the cost of a branch when the prediction is correct.
+  Scaled64 getPredictedPathCost(Scaled64 TrueCost, Scaled64 FalseCost,
+                                const SelectInst *SI);
+
   // Returns true if the target architecture supports lowering a given select.
   bool isSelectKindSupported(SelectInst *SI);
 };
 } // namespace
 
 char SelectOptimize::ID = 0;
 
 INITIALIZE_PASS_BEGIN(SelectOptimize, DEBUG_TYPE, "Optimize selects", false,
bool SelectOptimize::runOnFunction(Function &F) {
 
   TTI = &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
   DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
   LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
   BPI.reset(new BranchProbabilityInfo(F, *LI));
   BFI.reset(new BlockFrequencyInfo(F, *BPI, *LI));
   PSI = &getAnalysis<ProfileSummaryInfoWrapperPass>().getPSI();
   ORE = &getAnalysis<OptimizationRemarkEmitterWrapperPass>().getORE();
+  TSchedModel.init(TSI);
 
   // When optimizing for size, selects are preferable over branches.
   if (F.hasOptSize() || llvm::shouldOptimizeForSize(&F, PSI, BFI.get()))
     return false;
 
   return optimizeSelects(F);
 }
 
for (BasicBlock &BB : F) {
     collectSelectGroups(BB, SIGroups);
   }
 
   // Determine for which select groups it is profitable converting to branches.
   findProfitableSIGroupsBase(SIGroups, ProfSIGroups);
 }
 
 void SelectOptimize::optimizeSelectsInnerLoops(Function &F,
-                                               SelectGroups &ProfSIGroups) {}
+                                               SelectGroups &ProfSIGroups) {
+  SmallVector<Loop *, 4> Loops(LI->begin(), LI->end());
+  // Need to check size on each iteration as we accumulate child loops.
+  for (unsigned long i = 0; i < Loops.size(); ++i)
+    for (Loop *ChildL : Loops[i]->getSubLoops())
+      Loops.push_back(ChildL);
+
+  for (Loop *L : Loops) {
+    if (!L->isInnermost())
+      continue;
+
+    SelectGroups SIGroups;
+    for (BasicBlock *BB : L->getBlocks())
+      collectSelectGroups(*BB, SIGroups);
+
+    findProfitableSIGroupsInnerLoops(L, SIGroups, ProfSIGroups);
+  }
+}
 
 /// If \p isTrue is true, return the true value of \p SI, otherwise return
 /// false value of \p SI. If the true/false value of \p SI is defined by any
 /// select instructions in \p Selects, look through the defining select
 /// instruction until the true/false value is not defined in \p Selects.
 static Value *
 getTrueOrFalseValue(SelectInst *SI, bool isTrue,
                     const SmallPtrSet<const Instruction *, 2> &Selects) {
void SelectOptimize::findProfitableSIGroupsBase(SelectGroups &SIGroups,
                                                 SelectGroups &ProfSIGroups) {
   for (SelectGroup &ASI : SIGroups) {
     ++NumSelectOptAnalyzed;
     if (isConvertToBranchProfitableBase(ASI))
       ProfSIGroups.push_back(ASI);
   }
 }
 
+void SelectOptimize::findProfitableSIGroupsInnerLoops(
+    const Loop *L, SelectGroups &SIGroups, SelectGroups &ProfSIGroups) {
+  NumSelectOptAnalyzed += SIGroups.size();
+  // For each select group in an inner-most loop,
+  // a branch is more preferable than a select/conditional-move if:
+  // i) conversion to branches for all the select groups of the loop satisfies
+  //    loop-level heuristics including reducing the loop's critical path by
+  //    some threshold (see SelectOptimize::checkLoopHeuristics); and
+  // ii) the total cost of the select group is cheaper with a branch compared
+  //     to its predicated version. The cost is in terms of latency and the cost
+  //     of a select group is the cost of its most expensive select instruction
+  //     (assuming infinite resources and thus fully leveraging available ILP).
+
+  DenseMap<const Instruction *, CostInfo> InstCostMap;
+  CostInfo LoopCost[2] = {{Scaled64::getZero(), Scaled64::getZero()},
+                          {Scaled64::getZero(), Scaled64::getZero()}};
+  if (!computeLoopCosts(L, SIGroups, InstCostMap, LoopCost) ||
+      !checkLoopHeuristics(L, LoopCost)) {
+    return;
+  }
+
+  for (SelectGroup &ASI : SIGroups) {
+    // Assuming infinite resources, the cost of a group of instructions is the
+    // cost of the most expensive instruction of the group.
+    Scaled64 SelectCost = Scaled64::getZero(), BranchCost = Scaled64::getZero();
+    for (SelectInst *SI : ASI) {
+      SelectCost = std::max(SelectCost, InstCostMap[SI].PredCost);
+      BranchCost = std::max(BranchCost, InstCostMap[SI].NonPredCost);
+    }
+    if (BranchCost < SelectCost) {
+      OptimizationRemark OR(DEBUG_TYPE, "SelectOpti", ASI.front());
+      OR << "Profitable to convert to branch (loop analysis). BranchCost="
+         << BranchCost.toString() << ", SelectCost=" << SelectCost.toString()
+         << ". ";
+      ORE->emit(OR);
+      ++NumSelectConvertedLoop;
+      ProfSIGroups.push_back(ASI);
+    } else {
+      OptimizationRemarkMissed ORmiss(DEBUG_TYPE, "SelectOpti", ASI.front());
+      ORmiss << "Select is more profitable (loop analysis). BranchCost="
+             << BranchCost.toString()
+             << ", SelectCost=" << SelectCost.toString() << ". ";
+      ORE->emit(ORmiss);
+    }
+  }
+}
+
 bool SelectOptimize::isConvertToBranchProfitableBase(
     const SmallVector<SelectInst *, 2> &ASI) {
   SelectInst *SI = ASI.front();
   OptimizationRemark OR(DEBUG_TYPE, "SelectOpti", SI);
   OptimizationRemarkMissed ORmiss(DEBUG_TYPE, "SelectOpti", SI);
 
   // Skip cold basic blocks. Better to optimize for size for cold blocks.
   if (PSI->isColdBlock(SI->getParent(), BFI.get())) {
if (Sum != 0) {
       auto Probability = BranchProbability::getBranchProbability(Max, Sum);
       if (Probability > TTI->getPredictableBranchThreshold())
         return true;
     }
   }
   return false;
 }
 
+bool SelectOptimize::checkLoopHeuristics(const Loop *L,
+                                         const CostInfo LoopCost[2]) {
+  // Loop-level checks to determine if a non-predicated version (with branches)
+  // of the loop is more profitable than its predicated version.
+  //
+  OptimizationRemarkMissed ORmissL(DEBUG_TYPE, "SelectOpti",
+                                   L->getHeader()->getFirstNonPHI());
+
+  if (LoopCost[0].NonPredCost > LoopCost[0].PredCost ||
+      LoopCost[1].NonPredCost >= LoopCost[1].PredCost) {
+    ORmissL << "No select conversion in the loop due to no reduction of loop's "
+               "critical path. ";
+    ORE->emit(ORmissL);
+    return false;
+  }
+
+  Scaled64 Gain[2] = {LoopCost[0].PredCost - LoopCost[0].NonPredCost,
+                      LoopCost[1].PredCost - LoopCost[1].NonPredCost};
+
+  // Profitably converting to branches need to reduce the loop's critical path
+  // by at least some threshold (absolute gain of GainCycleThreshold cycles and
+  // relative gain of 12.5%).
+  if (Gain[1] < Scaled64::get(GainCycleThreshold) ||
+      Gain[1] * Scaled64::get(GainRelativeThreshold) < LoopCost[1].PredCost) {
+    Scaled64 RelativeGain = Scaled64::get(100) * Gain[1] / LoopCost[1].PredCost;
+    ORmissL << "No select conversion in the loop due to small reduction of "
+               "loop's critical path. Gain="
+            << Gain[1].toString()
+            << ", RelativeGain=" << RelativeGain.toString() << "%. ";
+    ORE->emit(ORmissL);
+    return false;
+  }
+
+  // If the loop's critical path involves loop-carried dependences, the gradient
+  // of the gain needs to be at least GainGradientThreshold% (defaults to 25%).
+  // This check ensures that the latency reduction for the loop's critical path
+  // keeps decreasing with sufficient rate beyond the two analyzed loop
+  // iterations.
+  if (Gain[1] > Gain[0]) {
+    Scaled64 GradientGain = Scaled64::get(100) * (Gain[1] - Gain[0]) /
+                            (LoopCost[1].PredCost - LoopCost[0].PredCost);
+    if (GradientGain < Scaled64::get(GainGradientThreshold)) {
+      ORmissL << "No select conversion in the loop due to small gradient gain. "
+                 "GradientGain="
+              << GradientGain.toString() << "%. ";
+      ORE->emit(ORmissL);
+      return false;
+    }
+  }
+  // If the gain decreases it is not profitable to convert.
+  else if (Gain[1] < Gain[0]) {
+    ORmissL
+        << "No select conversion in the loop due to negative gradient gain. ";
+    ORE->emit(ORmissL);
+    return false;
+  }
+
+  // Non-predicated version of the loop is more profitable than its
+  // predicated version.
+  return true;
+}
+
+// Computes instruction and loop-critical-path costs for both the predicated
+// and non-predicated version of the given loop.
+// Returns false if unable to compute these costs due to invalid cost of loop
+// instruction(s).
+bool SelectOptimize::computeLoopCosts(
+    const Loop *L, const SelectGroups &SIGroups,
+    DenseMap<const Instruction *, CostInfo> &InstCostMap, CostInfo *LoopCost) {
+  const auto &SIset = getSIset(SIGroups);
+  // Compute instruction and loop-critical-path costs across two iterations for
+  // both predicated and non-predicated version.
+  const unsigned Iterations = 2;
+  for (unsigned Iter = 0; Iter < Iterations; ++Iter) {
+    // Cost of the loop's critical path.
+    CostInfo &MaxCost = LoopCost[Iter];
+    for (BasicBlock *BB : L->getBlocks()) {
+      for (const Instruction &I : *BB) {
+        if (I.isDebugOrPseudoInst())
+          continue;
+        // Compute the predicated and non-predicated cost of the instruction.
+        Scaled64 IPredCost = Scaled64::getZero(),
+                 INonPredCost = Scaled64::getZero();
+
+        // Assume infinite resources that allow to fully exploit the available
+        // instruction-level parallelism.
+        // InstCost = InstLatency + max(Op1Cost, Op2Cost, … OpNCost)
+        for (const Use &U : I.operands()) {
+          auto UI = dyn_cast<Instruction>(U.get());
+          if (!UI)
+            continue;
+          if (InstCostMap.count(UI)) {
+            IPredCost = std::max(IPredCost, InstCostMap[UI].PredCost);
+            INonPredCost = std::max(INonPredCost, InstCostMap[UI].NonPredCost);
+          }
+        }
+        auto ILatency = computeInstCost(&I);
+        if (!ILatency.hasValue()) {
+          OptimizationRemarkMissed ORmissL(DEBUG_TYPE, "SelectOpti", &I);
+          ORmissL << "Invalid instruction cost preventing analysis and "
+                     "optimization of the inner-most loop containing this "
+                     "instruction. ";
+          ORE->emit(ORmissL);
+          return false;
+        }
+        IPredCost += Scaled64::get(ILatency.getValue());
+        INonPredCost += Scaled64::get(ILatency.getValue());
+
+        // For a select that can be converted to branch,
+        // compute its cost as a branch (non-predicated cost).
+        //
+        // BranchCost = PredictedPathCost + MispredictCost
+        // PredictedPathCost = TrueOpCost * TrueProb + FalseOpCost * FalseProb
+        // MispredictCost = max(MispredictPenalty, CondCost) * MispredictRate
+        if (SIset.contains(&I)) {
+          auto SI = dyn_cast<SelectInst>(&I);
+
+          Scaled64 TrueOpCost = Scaled64::getZero(),
+                   FalseOpCost = Scaled64::getZero();
+          if (auto *TI = dyn_cast<Instruction>(SI->getTrueValue()))
+            if (InstCostMap.count(TI))
+              TrueOpCost = InstCostMap[TI].NonPredCost;
+          if (auto *FI = dyn_cast<Instruction>(SI->getFalseValue()))
+            if (InstCostMap.count(FI))
+              FalseOpCost = InstCostMap[FI].NonPredCost;
+          Scaled64 PredictedPathCost =
+              getPredictedPathCost(TrueOpCost, FalseOpCost, SI);
+
+          Scaled64 CondCost = Scaled64::getZero();
+          if (auto *CI = dyn_cast<Instruction>(SI->getCondition()))
+            if (InstCostMap.count(CI))
+              CondCost = InstCostMap[CI].NonPredCost;
+          Scaled64 MispredictCost = getMispredictionCost(SI, CondCost);
+
+          INonPredCost = PredictedPathCost + MispredictCost;
+        }
+
+        InstCostMap[&I] = {IPredCost, INonPredCost};
+        MaxCost.PredCost = std::max(MaxCost.PredCost, IPredCost);
+        MaxCost.NonPredCost = std::max(MaxCost.NonPredCost, INonPredCost);
+      }
+    }
+  }
+  return true;
+}
+
+SmallPtrSet<const Instruction *, 2>
+SelectOptimize::getSIset(const SelectGroups &SIGroups) {
+  SmallPtrSet<const Instruction *, 2> SIset;
+  for (const SelectGroup &ASI : SIGroups)
+    for (const SelectInst *SI : ASI)
+      SIset.insert(SI);
+  return SIset;
+}
+
+Optional<uint64_t> SelectOptimize::computeInstCost(const Instruction *I) {
+  InstructionCost ICost =
+      TTI->getInstructionCost(I, TargetTransformInfo::TCK_Latency);
+  if (auto OC = ICost.getValue())
+    return Optional<uint64_t>(OC.getValue());
+  return Optional<uint64_t>(None);
+}
+
+ScaledNumber<uint64_t>
+SelectOptimize::getMispredictionCost(const SelectInst *SI,
+                                     const Scaled64 CondCost) {
+  uint64_t MispredictPenalty = TSchedModel.getMCSchedModel()->MispredictPenalty;
+
+  // Account for the default misprediction rate when using a branch
+  // (conservatively set to 25% by default).
+  uint64_t MispredictRate = MispredictDefaultRate;
+  // If the select condition is obviously predictable, then the misprediction
+  // rate is zero.
+  if (isSelectHighlyPredictable(SI))
+    MispredictRate = 0;
+
+  // CondCost is included to account for cases where the computation of the
+  // condition is part of a long dependence chain (potentially loop-carried)
+  // that would delay detection of a misprediction and increase its cost.
+  Scaled64 MispredictCost =
+      std::max(Scaled64::get(MispredictPenalty), CondCost) *
+      Scaled64::get(MispredictRate);
+  MispredictCost /= Scaled64::get(100);
+
+  return MispredictCost;
+}
+
+// Returns the cost of a branch when the prediction is correct.
+// TrueCost * TrueProbability + FalseCost * FalseProbability.
+ScaledNumber<uint64_t>
+SelectOptimize::getPredictedPathCost(Scaled64 TrueCost, Scaled64 FalseCost,
+                                     const SelectInst *SI) {
+  Scaled64 PredPathCost;
+  uint64_t TrueWeight, FalseWeight;
+  if (SI->extractProfMetadata(TrueWeight, FalseWeight)) {
+    uint64_t SumWeight = TrueWeight + FalseWeight;
+    if (SumWeight != 0) {
+      PredPathCost = TrueCost * Scaled64::get(TrueWeight) +
+                     FalseCost * Scaled64::get(FalseWeight);
+      PredPathCost /= Scaled64::get(SumWeight);
+      return PredPathCost;
+    }
+  }
+  // Without branch weight metadata, we assume 75% for the one path and 25% for
+  // the other, and pick the result with the biggest cost.
+  PredPathCost = std::max(TrueCost * Scaled64::get(3) + FalseCost,
+                          FalseCost * Scaled64::get(3) + TrueCost);
+  PredPathCost /= Scaled64::get(4);
+  return PredPathCost;
+}
+
 bool SelectOptimize::isSelectKindSupported(SelectInst *SI) {
   bool VectorCond = !SI->getCondition()->getType()->isIntegerTy(1);
   if (VectorCond)
     return false;
   TargetLowering::SelectSupportKind SelectKind;
   if (SI->getType()->isVectorTy())
     SelectKind = TargetLowering::ScalarCondVectorVal;
   else
     SelectKind = TargetLowering::ScalarValSelect;
   return TLI->isSelectSupported(SelectKind);
 }

llvm/test/CodeGen/X86/select-optimize.ll

 ; CHECK-NEXT:    ret i32 [[ADD]]
 ;
   %load = load i32, i32* %b, align 4
   %sel = select i1 %cmp, i32 %x, i32 %load
   %add = add i32 %sel, %load
   ret i32 %add
 }
 
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;; Test loop heuristic: loop-level critical-path analysis
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+;; Use of cmov in this test would put a load and a fsub on the critical path.
+;; Loop-level analysis should decide to form a branch.
+;;
+;;double cmov_on_critical_path(int n, double x, double *a) {
+;;  for (int i = 0; i < n; i++) {
+;;    double r = a[i];
+;;    if (x > r)
+;;      // 50% of iterations
+;;      x -= r;
+;;  }
+;;  return x;
+;;}
+define double @cmov_on_critical_path(i32 %n, double %x, double* nocapture %a) {
+; CHECK-LABEL: @cmov_on_critical_path(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 0
+; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; CHECK:       for.cond.cleanup:
+; CHECK-NEXT:    ret double [[X:%.*]]
+; CHECK:       for.body.preheader:
+; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[SELECT_END:%.*]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[X1:%.*]] = phi double [ [[X2:%.*]], [[SELECT_END]] ], [ [[X]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, double* [[A:%.*]], i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[R:%.*]] = load double, double* [[ARRAYIDX]], align 8
+; CHECK-NEXT:    [[SUB:%.*]] = fsub double [[X1]], [[R]]
+; CHECK-NEXT:    [[CMP2:%.*]] = fcmp ogt double [[X1]], [[R]]
+; CHECK-NEXT:    [[X2_FROZEN:%.*]] = freeze i1 [[CMP2]]
+; CHECK-NEXT:    br i1 [[X2_FROZEN]], label [[SELECT_END]], label [[SELECT_FALSE:%.*]], !prof [[PROF27:![0-9]+]]
+; CHECK:       select.false:
+; CHECK-NEXT:    br label [[SELECT_END]]
+; CHECK:       select.end:
+; CHECK-NEXT:    [[X2]] = phi double [ [[SUB]], [[FOR_BODY]] ], [ [[X1]], [[SELECT_FALSE]] ]
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_EXIT:%.*]], label [[FOR_BODY]]
+; CHECK:       for.exit:
+; CHECK-NEXT:    ret double [[X2]]
+;
+entry:
+  %cmp1 = icmp sgt i32 %n, 0
+  br i1 %cmp1, label %for.body.preheader, label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %entry
+  ret double %x
+
+for.body.preheader:                               ; preds = %entry
+  %wide.trip.count = zext i32 %n to i64
+  br label %for.body
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
+  %x1 = phi double [ %x2, %for.body ], [ %x, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds double, double* %a, i64 %indvars.iv
+  %r = load double, double* %arrayidx, align 8
+  %sub = fsub double %x1, %r
+  %cmp2 = fcmp ogt double %x1, %r
+  %x2 = select i1 %cmp2, double %sub, double %x1, !prof !18
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count
+  br i1 %exitcond, label %for.exit, label %for.body
+
+for.exit:                                         ; preds = %for.body
+  ret double %x2
+}
+
+;; The common path includes expensive operations (load and fsub) making
+;; branch similarly expensive to cmov, and thus the gain is small.
+;; Loop-level analysis should decide on not forming a branch.
+;;
+;;double small_gain(int n, double x, double *a) {
+;;  for (int i = 0; i < n; i++) {
+;;    double r = a[i];
+;;    if (x > r)
+;;      // 99% of iterations
+;;      x -= r;
+;;  }
+;;  return x;
+;;}
+define double @small_gain(i32 %n, double %x, double* nocapture %a) {
+; CHECK-LABEL: @small_gain(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 0
+; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; CHECK:       for.cond.cleanup:
+; CHECK-NEXT:    ret double [[X:%.*]]
+; CHECK:       for.body.preheader:
+; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[X1:%.*]] = phi double [ [[X2:%.*]], [[FOR_BODY]] ], [ [[X]], [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, double* [[A:%.*]], i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[R:%.*]] = load double, double* [[ARRAYIDX]], align 8
+; CHECK-NEXT:    [[SUB:%.*]] = fsub double [[X1]], [[R]]
+; CHECK-NEXT:    [[CMP2:%.*]] = fcmp ole double [[X1]], [[R]]
+; CHECK-NEXT:    [[X2]] = select i1 [[CMP2]], double [[X1]], double [[SUB]], !prof [[PROF18]]
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_EXIT:%.*]], label [[FOR_BODY]]
+; CHECK:       for.exit:
+; CHECK-NEXT:    ret double [[X2]]
+;
+entry:
+  %cmp1 = icmp sgt i32 %n, 0
+  br i1 %cmp1, label %for.body.preheader, label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %entry
+  ret double %x
+
+for.body.preheader:                               ; preds = %entry
+  %wide.trip.count = zext i32 %n to i64
+  br label %for.body
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
+  %x1 = phi double [ %x2, %for.body ], [ %x, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds double, double* %a, i64 %indvars.iv
+  %r = load double, double* %arrayidx, align 8
+  %sub = fsub double %x1, %r
+  %cmp2 = fcmp ole double %x1, %r
+  %x2 = select i1 %cmp2, double %x1, double %sub, !prof !17
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count
+  br i1 %exitcond, label %for.exit, label %for.body
+
+for.exit:                                         ; preds = %for.body
+  ret double %x2
+}
+
+;; Use of a branch in this test would avoid executing a load and several
+;; floating-point operations for most cases (70% of the time).
+;; Yet, the gain is not increasing much per iteration (small gradient gain).
+;; Loop-level analysis should decide not to form a branch.
+;;
+;;double small_gradient(int n, double x, double *a) {
+;;  for (int i = 0; i < n; i++) {
+;;    double r = 2 * a[i] + i;
+;;    if (r > 0)
+;;      // 30% of iterations
+;;      x -= r;
+;;  }
+;;  return x;
+;;}
+define double @small_gradient(i32 %n, double %x, double* nocapture %a) {
+; CHECK-LABEL: @small_gradient(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CMP8:%.*]] = icmp sgt i32 [[N:%.*]], 0
+; CHECK-NEXT:    br i1 [[CMP8]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; CHECK:       for.body.preheader:
+; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.cond.cleanup:
+; CHECK-NEXT:    [[X_ADDR_0_LCSSA:%.*]] = phi double [ [[X:%.*]], [[ENTRY:%.*]] ], [ [[X_ADDR_1:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    ret double [[X_ADDR_0_LCSSA]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[X_ADDR_010:%.*]] = phi double [ [[X]], [[FOR_BODY_PREHEADER]] ], [ [[X_ADDR_1]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, double* [[A:%.*]], i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[TMP0:%.*]] = load double, double* [[ARRAYIDX]], align 8
+; CHECK-NEXT:    [[TMP1:%.*]] = call double @llvm.fmuladd.f64(double [[TMP0]], double 2.000000e+00, double 1.000000e+00)
+; CHECK-NEXT:    [[CMP1:%.*]] = fcmp ogt double [[TMP1]], 0.000000e+00
+; CHECK-NEXT:    [[SUB:%.*]] = select i1 [[CMP1]], double [[TMP1]], double 0.000000e+00, !prof [[PROF28:![0-9]+]]
+; CHECK-NEXT:    [[X_ADDR_1]] = fsub double [[X_ADDR_010]], [[SUB]]
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]]
+;
+entry:
+  %cmp8 = icmp sgt i32 %n, 0
+  br i1 %cmp8, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:                               ; preds = %entry
+  %wide.trip.count = zext i32 %n to i64
+  br label %for.body
+
+for.cond.cleanup:                                 ; preds = %for.body, %entry
+  %x.addr.0.lcssa = phi double [ %x, %entry ], [ %x.addr.1, %for.body ]
+  ret double %x.addr.0.lcssa
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
+  %x.addr.010 = phi double [ %x, %for.body.preheader ], [ %x.addr.1, %for.body ]
+  %arrayidx = getelementptr inbounds double, double* %a, i64 %indvars.iv
+  %0 = load double, double* %arrayidx, align 8
+  %1 = call double @llvm.fmuladd.f64(double %0, double 2.000000e+00, double 1.000000e+00)
+  %cmp1 = fcmp ogt double %1, 0.000000e+00
+  %sub = select i1 %cmp1, double %1, double 0.000000e+00, !prof !28
+  %x.addr.1 = fsub double %x.addr.010, %sub
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
+  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+;; One select on the critical path and one off the critical path.
+;; Loop-level analysis should decide to form a branch only for
+;; the select on the critical path.
+;;
+;;double loop_select_groups(int n, double x, double *a, int k) {
+;;  int c = 0;
+;;  for (int i = 0; i < n; i++) {
+;;    double r = a[i];
+;;    if (x > r)
+;;      x -= r;
+;;    if (i == k)
+;;      c += n;
+;;  }
+;;  return x + c;
+;;}
+define double @loop_select_groups(i32 %n, double %x, double* nocapture %a, i32 %k) {
+; CHECK-LABEL: @loop_select_groups(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CMP19:%.*]] = icmp sgt i32 [[N:%.*]], 0
+; CHECK-NEXT:    br i1 [[CMP19]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; CHECK:       for.body.preheader:
+; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.cond.cleanup.loopexit:
+; CHECK-NEXT:    [[PHI_CAST:%.*]] = sitofp i32 [[C_1:%.*]] to double
+; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
+; CHECK:       for.cond.cleanup:
+; CHECK-NEXT:    [[C_0_LCSSA:%.*]] = phi double [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[PHI_CAST]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
+; CHECK-NEXT:    [[X_ADDR_0_LCSSA:%.*]] = phi double [ [[X:%.*]], [[ENTRY]] ], [ [[X_ADDR_1:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT]] ]
+; CHECK-NEXT:    [[ADD5:%.*]] = fadd double [[X_ADDR_0_LCSSA]], [[C_0_LCSSA]]
+; CHECK-NEXT:    ret double [[ADD5]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[SELECT_END:%.*]] ]
+; CHECK-NEXT:    [[X_ADDR_022:%.*]] = phi double [ [[X]], [[FOR_BODY_PREHEADER]] ], [ [[X_ADDR_1]], [[SELECT_END]] ]
+; CHECK-NEXT:    [[C_020:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[C_1]], [[SELECT_END]] ]
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, double* [[A:%.*]], i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[TMP0:%.*]] = load double, double* [[ARRAYIDX]], align 8
+; CHECK-NEXT:    [[CMP1:%.*]] = fcmp ogt double [[X_ADDR_022]], [[TMP0]]
+; CHECK-NEXT:    [[SUB_FROZEN:%.*]] = freeze i1 [[CMP1]]
+; CHECK-NEXT:    br i1 [[SUB_FROZEN]], label [[SELECT_END]], label [[SELECT_FALSE:%.*]]
+; CHECK:       select.false:
+; CHECK-NEXT:    br label [[SELECT_END]]
+; CHECK:       select.end:
+; CHECK-NEXT:    [[SUB:%.*]] = phi double [ [[TMP0]], [[FOR_BODY]] ], [ 0.000000e+00, [[SELECT_FALSE]] ]
+; CHECK-NEXT:    [[X_ADDR_1]] = fsub double [[X_ADDR_022]], [[SUB]]
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[INDVARS_IV]] to i32
+; CHECK-NEXT:    [[CMP2:%.*]] = icmp eq i32 [[K:%.*]], [[N]]
+; CHECK-NEXT:    [[ADD:%.*]] = select i1 [[CMP2]], i32 [[N]], i32 0
+; CHECK-NEXT:    [[C_1]] = add nsw i32 [[ADD]], [[C_020]]
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
+;
+entry:
+  %cmp19 = icmp sgt i32 %n, 0
+  br i1 %cmp19, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:                               ; preds = %entry
+  %wide.trip.count = zext i32 %n to i64
+  br label %for.body
+
+for.cond.cleanup.loopexit:                        ; preds = %for.body
+  %phi.cast = sitofp i32 %c.1 to double
+  br label %for.cond.cleanup
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  %c.0.lcssa = phi double [ 0.000000e+00, %entry ], [ %phi.cast, %for.cond.cleanup.loopexit ]
+  %x.addr.0.lcssa = phi double [ %x, %entry ], [ %x.addr.1, %for.cond.cleanup.loopexit ]
+  %add5 = fadd double %x.addr.0.lcssa, %c.0.lcssa
+  ret double %add5
+
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
+  %x.addr.022 = phi double [ %x, %for.body.preheader ], [ %x.addr.1, %for.body ]
+  %c.020 = phi i32 [ 0, %for.body.preheader ], [ %c.1, %for.body ]
+  %arrayidx = getelementptr inbounds double, double* %a, i64 %indvars.iv
+  %0 = load double, double* %arrayidx, align 8
+  %cmp1 = fcmp ogt double %x.addr.022, %0
+  %sub = select i1 %cmp1, double %0, double 0.000000e+00
+  %x.addr.1 = fsub double %x.addr.022, %sub
+  %1 = trunc i64 %indvars.iv to i32
+  %cmp2 = icmp eq i32 %k, %n
+  %add = select i1 %cmp2, i32 %n, i32 0
+  %c.1 = add nsw i32 %add, %c.020
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
+  br i1 %exitcond.not, label %for.cond.cleanup.loopexit, label %for.body
+}
+
 ; Function Attrs: nounwind readnone speculatable willreturn
 declare void @llvm.dbg.value(metadata, metadata, metadata)
 
+; Function Attrs: mustprogress nofree nosync nounwind readnone speculatable willreturn
+declare double @llvm.fmuladd.f64(double, double, double)
+
 !llvm.module.flags = !{!0, !26, !27}
 !0 = !{i32 1, !"ProfileSummary", !1}
 !1 = !{!2, !3, !4, !5, !6, !7, !8, !9}
 !2 = !{!"ProfileFormat", !"InstrProf"}
 !3 = !{!"TotalCount", i64 10000}
 !4 = !{!"MaxCount", i64 10}
 !5 = !{!"MaxInternalCount", i64 1}
 !6 = !{!"MaxFunctionCount", i64 1000}

 !20 = !{}
 !21 = !DIFile(filename: "test.c", directory: "/test")
 !22 = distinct !DICompileUnit(language: DW_LANG_C99, file: !21, producer: "clang version 15.0.0", isOptimized: true, emissionKind: FullDebug, globals: !25, splitDebugInlining: false, nameTableKind: None)
 !23 = distinct !DISubprogram(name: "test", scope: !21, file: !21, line: 1, unit: !22)
 !24 = !DILocalVariable(name: "x", scope: !23)
 !25 = !{}
 !26 = !{i32 2, !"Dwarf Version", i32 4}
 !27 = !{i32 1, !"Debug Info Version", i32 3}
+!28 = !{!"branch_weights", i32 30, i32 70}