Use profile summary to disable peeling for huge working sets

[tejohnson]

Summary

Detect when the working set size of a profiled application is huge,
by comparing the number of counts required to reach the hot percentile
in the profile summary to a large threshold*.

When the working set size is determined to be huge, disable peeling
to avoid bloating the working set further.

*Note that the selected threshold (15K) is significantly larger than the
largest working set value in SPEC cpu2006 (which is gcc at around 11K).

[eraman]

lib/Analysis/ProfileSummaryInfo.cpp

69

The getMinCountForPercentile and getNumCountsForPercentile could be merged to a getEntryForPercentile

[davidxl]

Have we considered limiting the max number of peeled iterations instead of disabling it?

[tejohnson]

In D36288#831144, @davidxl wrote:

Have we considered limiting the max number of peeled iterations instead of disabling it?

I am planning to do some follow on tuning that would limit the factor based on additional profile info, but as a first heuristic disabling it helps quite a bit.

lib/Analysis/ProfileSummaryInfo.cpp

69

That's a good idea, will do so.

[tejohnson]

Implement Easwaran's suggestion

[davidxl]

lgtm

[mehdi_amini]

When the working set size is determined to be huge, disable peeling to avoid bloating the working set further.

Just lurking and curious: how is "working set" defined first? And then how is peeling bloating it?

include/llvm/Analysis/ProfileSummaryInfo.h

 class ProfileSummaryInfo {
 private:
   Module &M;
   std::unique_ptr<ProfileSummary> Summary;
   bool computeSummary();
   void computeThresholds();
   // Count thresholds to answer isHotCount and isColdCount queries.
   Optional<uint64_t> HotCountThreshold, ColdCountThreshold;
+  // True if the working set size of the code is considered huge,
+  // because the number of profile counts required to reach the hot
+  // percentile is above a huge threshold.
+  Optional<bool> HasHugeWorkingSetSize;
 
 public:
   ProfileSummaryInfo(Module &M) : M(M) {}
   ProfileSummaryInfo(ProfileSummaryInfo &&Arg)
       : M(Arg.M), Summary(std::move(Arg.Summary)) {}
 
   /// \brief Returns true if profile summary is available.
   bool hasProfileSummary() { return computeSummary(); }
public:
   bool invalidate(Module &, const PreservedAnalyses &,
                   ModuleAnalysisManager::Invalidator &) {
     return false;
   }
 
   /// Returns the profile count for \p CallInst.
   Optional<uint64_t> getProfileCount(const Instruction *CallInst,
                                      BlockFrequencyInfo *BFI);
+  /// Returns true if the working set size of the code is considered huge.
+  bool hasHugeWorkingSetSize();
   /// \brief Returns true if \p F has hot function entry.
   bool isFunctionEntryHot(const Function *F);
   /// Returns true if \p F has hot function entry or hot call edge.
   bool isFunctionHotInCallGraph(const Function *F);
   /// \brief Returns true if \p F has cold function entry.
   bool isFunctionEntryCold(const Function *F);
   /// Returns true if \p F has cold function entry or cold call edge.
   bool isFunctionColdInCallGraph(const Function *F);

lib/Analysis/ProfileSummaryInfo.cpp

static cl::opt<int> ProfileSummaryCutoffCold(
     cl::desc("A count is cold if it is below the minimum count"
              " to reach this percentile of total counts."));
 
 static cl::opt<bool> AccurateSampleProfile(
     "accurate-sample-profile", cl::Hidden, cl::init(false),
     cl::desc("If the sample profile is accurate, we will mark all un-sampled "
              "callsite as cold. Otherwise, treat un-sampled callsites as if "
              "we have no profile."));
+static cl::opt<unsigned> ProfileSummaryHugeWorkingSetSizeThreshold(
+    "profile-summary-huge-working-set-size-threshold", cl::Hidden,
+    cl::init(15000), cl::ZeroOrMore,
+    cl::desc("The code working set size is considered huge if the number of"
+             " blocks required to reach the -profile-summary-cutoff-hot"
+             " percentile exceeds this count."));
 
-// Find the minimum count to reach a desired percentile of counts.
-static uint64_t getMinCountForPercentile(SummaryEntryVector &DS,
+// Find the summary entry for a desired percentile of counts.
+static const ProfileSummaryEntry &getEntryForPercentile(SummaryEntryVector &DS,
                                                         uint64_t Percentile) {
   auto Compare = [](const ProfileSummaryEntry &Entry, uint64_t Percentile) {
     return Entry.Cutoff < Percentile;
   };
   auto It = std::lower_bound(DS.begin(), DS.end(), Percentile, Compare);
   // The required percentile has to be <= one of the percentiles in the
   // detailed summary.
   if (It == DS.end())
     report_fatal_error("Desired percentile exceeds the maximum cutoff");
-  return It->MinCount;
+  return *It;
 }
 
 // The profile summary metadata may be attached either by the frontend or by
 // any backend passes (IR level instrumentation, for example). This method
 // checks if the Summary is null and if so checks if the summary metadata is now
 // available in the module and parses it to get the Summary object. Returns true
 // if a valid Summary is available.
 bool ProfileSummaryInfo::computeSummary() {
   if (Summary)
     return true;
   auto *SummaryMD = M.getProfileSummary();
   if (!SummaryMD)
bool ProfileSummaryInfo::isFunctionEntryCold(const Function *F) {
   return FunctionCount && isColdCount(FunctionCount.getValue());
 }
 
 /// Compute the hot and cold thresholds.
 void ProfileSummaryInfo::computeThresholds() {
   if (!computeSummary())
     return;
   auto &DetailedSummary = Summary->getDetailedSummary();
-  HotCountThreshold =
-      getMinCountForPercentile(DetailedSummary, ProfileSummaryCutoffHot);
-  ColdCountThreshold =
-      getMinCountForPercentile(DetailedSummary, ProfileSummaryCutoffCold);
+  auto &HotEntry =
+      getEntryForPercentile(DetailedSummary, ProfileSummaryCutoffHot);
+  HotCountThreshold = HotEntry.MinCount;
+  auto &ColdEntry =
+      getEntryForPercentile(DetailedSummary, ProfileSummaryCutoffCold);
+  ColdCountThreshold = ColdEntry.MinCount;
+  HasHugeWorkingSetSize =
+      HotEntry.NumCounts > ProfileSummaryHugeWorkingSetSizeThreshold;
+}
+
+bool ProfileSummaryInfo::hasHugeWorkingSetSize() {
+  if (!HasHugeWorkingSetSize)
+    computeThresholds();
+  return HasHugeWorkingSetSize && HasHugeWorkingSetSize.getValue();
 }
 
 bool ProfileSummaryInfo::isHotCount(uint64_t C) {
   if (!HotCountThreshold)
     computeThresholds();
   return HotCountThreshold && C >= HotCountThreshold.getValue();
 }
 

lib/Transforms/Scalar/LoopUnrollPass.cpp

 #include "llvm/ADT/SetVector.h"
 #include "llvm/Analysis/AssumptionCache.h"
 #include "llvm/Analysis/CodeMetrics.h"
 #include "llvm/Analysis/GlobalsModRef.h"
 #include "llvm/Analysis/InstructionSimplify.h"
 #include "llvm/Analysis/LoopPass.h"
 #include "llvm/Analysis/LoopUnrollAnalyzer.h"
 #include "llvm/Analysis/OptimizationDiagnosticInfo.h"
+#include "llvm/Analysis/ProfileSummaryInfo.h"
 #include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/InstVisitor.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Metadata.h"
 #include "llvm/Support/CommandLine.h"
PreservedAnalyses LoopUnrollPass::run(Function &F,
                                       FunctionAnalysisManager &AM) {
   auto &SE = AM.getResult<ScalarEvolutionAnalysis>(F);
   auto &LI = AM.getResult<LoopAnalysis>(F);
   auto &TTI = AM.getResult<TargetIRAnalysis>(F);
   auto &DT = AM.getResult<DominatorTreeAnalysis>(F);
   auto &AC = AM.getResult<AssumptionAnalysis>(F);
   auto &ORE = AM.getResult<OptimizationRemarkEmitterAnalysis>(F);
 
+  const ModuleAnalysisManager &MAM =
+      AM.getResult<ModuleAnalysisManagerFunctionProxy>(F).getManager();
+  ProfileSummaryInfo *PSI =
+      MAM.getCachedResult<ProfileSummaryAnalysis>(*F.getParent());
+
   bool Changed = false;
 
   // The unroller requires loops to be in simplified form, and also needs LCSSA.
   // Since simplification may add new inner loops, it has to run before the
   // legality and profitability checks. This means running the loop unroller
   // will simplify all loops, regardless of whether anything end up being
   // unrolled.
   for (auto &L : LI) {

 #ifndef NDEBUG
     Loop *ParentL = L.getParentLoop();
 #endif
 
     // The API here is quite complex to call, but there are only two interesting
     // states we support: partial and full (or "simple") unrolling. However, to
     // enable these things we actually pass "None" in for the optional to avoid
     // providing an explicit choice.
-    Optional<bool> AllowPartialParam, RuntimeParam, UpperBoundParam;
-    bool CurChanged = tryToUnrollLoop(
-        &L, DT, &LI, SE, TTI, AC, ORE,
+    Optional<bool> AllowPartialParam, RuntimeParam, UpperBoundParam,
+        AllowPeeling;
+    // Check if the profile summary indicates that the profiled application
+    // has a huge working set size, in which case we disable peeling to avoid
+    // bloating it further.
+    if (PSI && PSI->hasHugeWorkingSetSize())
+      AllowPeeling = false;
+    bool CurChanged =
+        tryToUnrollLoop(&L, DT, &LI, SE, TTI, AC, ORE,
                         /*PreserveLCSSA*/ true, OptLevel, /*Count*/ None,
-        /*Threshold*/ None, AllowPartialParam, RuntimeParam, UpperBoundParam,
-        /*AllowPeeling*/ None);
+                        /*Threshold*/ None, AllowPartialParam, RuntimeParam,
+                        UpperBoundParam, AllowPeeling);
     Changed |= CurChanged;
 
     // The parent must not be damaged by unrolling!
 #ifndef NDEBUG
     if (CurChanged && ParentL)
       ParentL->verifyLoop();
 #endif
   }
 
   if (!Changed)
     return PreservedAnalyses::all();
 
   return getLoopPassPreservedAnalyses();
 }

test/Other/new-pm-defaults.ll

 ; CHECK-O-NEXT: Running analysis: BranchProbabilityAnalysis
 ; CHECK-O-NEXT: Running pass: LoopLoadEliminationPass
 ; CHECK-O-NEXT: Running analysis: LoopAccessAnalysis
 ; CHECK-O-NEXT: Running pass: InstCombinePass
 ; CHECK-O-NEXT: Running pass: SLPVectorizerPass
 ; CHECK-O-NEXT: Running pass: SimplifyCFGPass
 ; CHECK-O-NEXT: Running pass: InstCombinePass
 ; CHECK-O-NEXT: Running pass: LoopUnrollPass
+; CHECK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
 ; CHECK-O-NEXT: Running pass: InstCombinePass
 ; CHECK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}OptimizationRemarkEmitterAnalysis
 ; CHECK-O-NEXT: Running pass: FunctionToLoopPassAdaptor<{{.*}}LICMPass
 ; CHECK-O-NEXT: Running pass: AlignmentFromAssumptionsPass
 ; CHECK-O-NEXT: Running pass: LoopSinkPass
 ; CHECK-O-NEXT: Running pass: InstSimplifierPass
 ; CHECK-O-NEXT: Running pass: SimplifyCFGPass
 ; CHECK-O-NEXT: Finished llvm::Function pass manager run.

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

 ; CHECK-POSTLINK-O-NEXT: Running analysis: BranchProbabilityAnalysis
 ; CHECK-POSTLINK-O-NEXT: Running pass: LoopLoadEliminationPass
 ; CHECK-POSTLINK-O-NEXT: Running analysis: LoopAccessAnalysis
 ; CHECK-POSTLINK-O-NEXT: Running pass: InstCombinePass
 ; CHECK-POSTLINK-O-NEXT: Running pass: SLPVectorizerPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: SimplifyCFGPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: InstCombinePass
 ; CHECK-POSTLINK-O-NEXT: Running pass: LoopUnrollPass
+; CHECK-POSTLINK-O-NEXT: Running analysis: OuterAnalysisManagerProxy
 ; CHECK-POSTLINK-O-NEXT: Running pass: InstCombinePass
 ; CHECK-POSTLINK-O-NEXT: Running pass: RequireAnalysisPass<{{.*}}OptimizationRemarkEmitterAnalysis
 ; CHECK-POSTLINK-O-NEXT: Running pass: FunctionToLoopPassAdaptor<{{.*}}LICMPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: AlignmentFromAssumptionsPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: LoopSinkPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: InstSimplifierPass
 ; CHECK-POSTLINK-O-NEXT: Running pass: SimplifyCFGPass
 ; CHECK-POSTLINK-O-NEXT: Finished llvm::Function pass manager run.

test/Transforms/LoopUnroll/peel-loop-pgo.ll

 ; RUN: opt < %s -S -debug-only=loop-unroll -loop-unroll 2>&1 | FileCheck %s
-; RUN: opt < %s -S -debug-only=loop-unroll -passes='require<opt-remark-emit>,unroll' 2>&1 | FileCheck %s
+; RUN: opt < %s -S -debug-only=loop-unroll -passes='require<profile-summary>,function(require<opt-remark-emit>,unroll)' 2>&1 | FileCheck %s
+; Confirm that peeling is disabled if the number of counts required to reach
+; the hot percentile is above the threshold.
+; RUN: opt < %s -S -profile-summary-huge-working-set-size-threshold=9 -debug-only=loop-unroll -passes='require<profile-summary>,function(require<opt-remark-emit>,unroll)' 2>&1 | FileCheck %s --check-prefix=NOPEEL
 ; REQUIRES: asserts
 
 ; Make sure we use the profile information correctly to peel-off 3 iterations
 ; from the loop, and update the branch weights for the peeled loop properly.
 
 ; CHECK: Loop Unroll: F[basic]
 ; CHECK: PEELING loop %for.body with iteration count 3!
 ; CHECK: Loop Unroll: F[optsize]
 ; CHECK-NOT: PEELING
 
 ; Confirm that no peeling occurs when we are performing full unrolling.
-; RUN: opt < %s -S -debug-only=loop-unroll -passes='require<opt-remark-emit>,loop(unroll-full)' 2>&1 | FileCheck %s --check-prefix=FULLUNROLL
-; FULLUNROLL-NOT: PEELING
+; RUN: opt < %s -S -debug-only=loop-unroll -passes='require<opt-remark-emit>,loop(unroll-full)' 2>&1 | FileCheck %s --check-prefix=NOPEEL
+; NOPEEL-NOT: PEELING
 
 ; CHECK-LABEL: @basic
-; CHECK: br i1 %{{.*}}, label %[[NEXT0:.*]], label %for.cond.for.end_crit_edge, !prof !1
+; CHECK: br i1 %{{.*}}, label %[[NEXT0:.*]], label %for.cond.for.end_crit_edge, !prof !15
 ; CHECK: [[NEXT0]]:
-; CHECK: br i1 %{{.*}}, label %[[NEXT1:.*]], label %for.cond.for.end_crit_edge, !prof !2
+; CHECK: br i1 %{{.*}}, label %[[NEXT1:.*]], label %for.cond.for.end_crit_edge, !prof !16
 ; CHECK: [[NEXT1]]:
-; CHECK: br i1 %{{.*}}, label %[[NEXT2:.*]], label %for.cond.for.end_crit_edge, !prof !3
+; CHECK: br i1 %{{.*}}, label %[[NEXT2:.*]], label %for.cond.for.end_crit_edge, !prof !17
 ; CHECK: [[NEXT2]]:
-; CHECK: br i1 %{{.*}}, label %for.body, label %{{.*}}, !prof !4
+; CHECK: br i1 %{{.*}}, label %for.body, label %{{.*}}, !prof !18
 
-define void @basic(i32* %p, i32 %k) #0 !prof !0 {
+define void @basic(i32* %p, i32 %k) #0 !prof !15 {
 entry:
   %cmp3 = icmp slt i32 0, %k
   br i1 %cmp3, label %for.body.lr.ph, label %for.end
 
 for.body.lr.ph:                                   ; preds = %entry
   br label %for.body
 
 for.body:                                         ; preds = %for.body.lr.ph, %for.body
   %i.05 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
   %p.addr.04 = phi i32* [ %p, %for.body.lr.ph ], [ %incdec.ptr, %for.body ]
   %incdec.ptr = getelementptr inbounds i32, i32* %p.addr.04, i32 1
   store i32 %i.05, i32* %p.addr.04, align 4
   %inc = add nsw i32 %i.05, 1
   %cmp = icmp slt i32 %inc, %k
-  br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge, !prof !1
+  br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge, !prof !16
 
 for.cond.for.end_crit_edge:                       ; preds = %for.body
   br label %for.end
 
 for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry
   ret void
 }
 
 ; We don't want to peel loops when optimizing for size.
 ; CHECK-LABEL: @optsize
 ; CHECK: for.body.lr.ph:
 ; CHECK-NEXT: br label %for.body
 ; CHECK: for.body:
 ; CHECK-NOT: br
 ; CHECK: br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge
-define void @optsize(i32* %p, i32 %k) #1 !prof !0 {
+define void @optsize(i32* %p, i32 %k) #1 !prof !15 {
 entry:
   %cmp3 = icmp slt i32 0, %k
   br i1 %cmp3, label %for.body.lr.ph, label %for.end
 
 for.body.lr.ph:                                   ; preds = %entry
   br label %for.body
 
 for.body:                                         ; preds = %for.body.lr.ph, %for.body
   %i.05 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
   %p.addr.04 = phi i32* [ %p, %for.body.lr.ph ], [ %incdec.ptr, %for.body ]
   %incdec.ptr = getelementptr inbounds i32, i32* %p.addr.04, i32 1
   store i32 %i.05, i32* %p.addr.04, align 4
   %inc = add nsw i32 %i.05, 1
   %cmp = icmp slt i32 %inc, %k
-  br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge, !prof !1
+  br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge, !prof !16
 
 for.cond.for.end_crit_edge:                       ; preds = %for.body
   br label %for.end
 
 for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry
   ret void
 }
 
 attributes #0 = { nounwind }
 attributes #1 = { nounwind optsize }
 
-!0 = !{!"function_entry_count", i64 1}
-!1 = !{!"branch_weights", i32 3001, i32 1001}
+!llvm.module.flags = !{!1}
 
-;CHECK: !1 = !{!"branch_weights", i32 900, i32 101}
-;CHECK: !2 = !{!"branch_weights", i32 540, i32 360}
-;CHECK: !3 = !{!"branch_weights", i32 162, i32 378}
-;CHECK: !4 = !{!"branch_weights", i32 1399, i32 162}
+!1 = !{i32 1, !"ProfileSummary", !2}
+!2 = !{!3, !4, !5, !6, !7, !8, !9, !10}
+!3 = !{!"ProfileFormat", !"InstrProf"}
+!4 = !{!"TotalCount", i64 10}
+!5 = !{!"MaxCount", i64 3}
+!6 = !{!"MaxInternalCount", i64 1}
+!7 = !{!"MaxFunctionCount", i64 3}
+!8 = !{!"NumCounts", i64 2}
+!9 = !{!"NumFunctions", i64 2}
+!10 = !{!"DetailedSummary", !11}
+!11 = !{!12, !13, !14}
+!12 = !{i32 10000, i64 3, i32 2}
+!13 = !{i32 999000, i64 1, i32 10}
+!14 = !{i32 999999, i64 1, i32 10}
+!15 = !{!"function_entry_count", i64 1}
+!16 = !{!"branch_weights", i32 3001, i32 1001}
+
+;CHECK: !15 = !{!"branch_weights", i32 900, i32 101}
+;CHECK: !16 = !{!"branch_weights", i32 540, i32 360}
+;CHECK: !17 = !{!"branch_weights", i32 162, i32 378}
+;CHECK: !18 = !{!"branch_weights", i32 1399, i32 162}