[AutoFDO] Inline replay for cold/small callees from sample profile loader

[wenlei]

Summary

Sample profile loader of AutoFDO tries to replay previous inlining using context sensitive profile. The replay only repeats inlining if the call site block is hot. As a result it punts inlining of small functions, some of which can be beneficial for size, and will still be inlined by CSGCC inliner later. The oscillation between sample profile loader's inlining and regular CGSSC inlining cause unnecessary loss of context-sensitive profile. It doesn't have much impact for inline decision itself, but it negatively affects post-inline profile quality as CGSCC inliner have to scale counts which is not as accurate as the original context sensitive profile, and bad post-inline profile can misguide code layout.

This change added regular Inline Cost calculation for sample profile loader, so we can inline small functions upfront under switch -sample-profile-inline-size. In addition -sample-profile-cold-inline-threshold is added so we can tune the separate size threshold - currently the default is chosen to be the same as regular inliner's cold call-site threshold. Last, -sample-profile-reinline-all is added to replay all inlining from previous compilation - this is for tuning only and the purpose is to avoid loss of context-sensitivity due to inlining churn at the cost of not using context-sensitive profile to drive inlining.

[wenlei]

update test case

[wmi]

Did performance test and I saw 0.4% improvement in an internal benchmark. That is a good improvement, thanks for the change!

llvm/lib/Transforms/IPO/SampleProfile.cpp

976–979

Can we inline all by setting sample-profile-cold-inline-threshold to a very large number so ProfileReInlineAll is not needed?

[wenlei]

address feedback

[wenlei]

Did performance test and I saw 0.4% improvement in an internal benchmark. That is a good improvement, thanks for the change!

Thanks for doing the measurement. I'm really glad that this change and other recent ones in this area helped your workload too. These are like "prototype" for the stuff we talked about a while ago. These numbers are important signals so we know what we're trying to do is at least a step in the right direction.

llvm/lib/Transforms/IPO/SampleProfile.cpp

976–979

good suggestion. changed.

[wmi]

LGTM. Thanks!

llvm/lib/Transforms/IPO/SampleProfile.cpp

static cl::opt<bool> ProfileMergeInlinee(
     cl::desc("Merge past inlinee's profile to outline version if sample "
              "profile loader decided not to inline a call site."));
 
 static cl::opt<bool> ProfileTopDownLoad(
     "sample-profile-top-down-load", cl::Hidden, cl::init(false),
     cl::desc("Do profile annotation and inlining for functions in top-down "
              "order of call graph during sample profile loading."));
 
+static cl::opt<bool> ProfileSizeInline(
+    "sample-profile-inline-size", cl::Hidden, cl::init(false),
+    cl::desc("Inline cold call sites in profile loader if it's beneficial "
+             "for code size."));
+
+static cl::opt<int> SampleColdCallSiteThreshold(
+    "sample-profile-cold-inline-threshold", cl::Hidden, cl::init(45),
+    cl::desc("Threshold for inlining cold callsites"));
+
 namespace {
 
 using BlockWeightMap = DenseMap<const BasicBlock *, uint64_t>;
 using EquivalenceClassMap = DenseMap<const BasicBlock *, const BasicBlock *>;
 using Edge = std::pair<const BasicBlock *, const BasicBlock *>;
 using EdgeWeightMap = DenseMap<Edge, uint64_t>;
 using BlockEdgeMap =
     DenseMap<const BasicBlock *, SmallVector<const BasicBlock *, 8>>;
protected:
   const FunctionSamples *findCalleeFunctionSamples(const Instruction &I) const;
   std::vector<const FunctionSamples *>
   findIndirectCallFunctionSamples(const Instruction &I, uint64_t &Sum) const;
   mutable DenseMap<const DILocation *, const FunctionSamples *> DILocation2SampleMap;
   const FunctionSamples *findFunctionSamples(const Instruction &I) const;
   bool inlineCallInstruction(Instruction *I);
   bool inlineHotFunctions(Function &F,
                           DenseSet<GlobalValue::GUID> &InlinedGUIDs);
+  // Inline cold/small functions in addition to hot ones
+  bool shouldInlineColdCallee(Instruction &CallInst);
   void printEdgeWeight(raw_ostream &OS, Edge E);
   void printBlockWeight(raw_ostream &OS, const BasicBlock *BB) const;
   void printBlockEquivalence(raw_ostream &OS, const BasicBlock *BB);
   bool computeBlockWeights(Function &F);
   void findEquivalenceClasses(Function &F);
   template <bool IsPostDom>
   void findEquivalencesFor(BasicBlock *BB1, ArrayRef<BasicBlock *> Descendants,
                            DominatorTreeBase<BasicBlock, IsPostDom> *DomTree);
if (InlineFunction(CS, IFI)) {
     ORE->emit(OptimizationRemark(DEBUG_TYPE, "HotInline", DLoc, BB)
               << "inlined hot callee '" << ore::NV("Callee", CalledFunction)
               << "' into '" << ore::NV("Caller", BB->getParent()) << "'");
     return true;
   }
   return false;
 }
 
+bool SampleProfileLoader::shouldInlineColdCallee(Instruction &CallInst) {
+  if (!ProfileSizeInline)
+    return false;
+
+  Function *Callee = CallSite(&CallInst).getCalledFunction();
+  if (Callee == nullptr)
+    return false;
+
+  InlineCost Cost =
+      getInlineCost(cast<CallBase>(CallInst), getInlineParams(),
+                    GetTTI(*Callee), GetAC, None, nullptr, nullptr);
+
+  return Cost.getCost() <= SampleColdCallSiteThreshold;
+}
+
 /// Iteratively inline hot callsites of a function.
 ///
 /// Iteratively traverse all callsites of the function \p F, and find if
 /// the corresponding inlined instance exists and is hot in profile. If
 /// it is hot enough, inline the callsites and adds new callsites of the
 /// callee into the caller. If the call is an indirect call, first promote
 /// it to direct call. Each indirect call is limited with a single target.
 ///
bool SampleProfileLoader::inlineHotFunctions(
 
   DenseMap<Instruction *, const FunctionSamples *> localNotInlinedCallSites;
   bool Changed = false;
   while (true) {
     bool LocalChanged = false;
     SmallVector<Instruction *, 10> CIS;
     for (auto &BB : F) {
       bool Hot = false;
-      SmallVector<Instruction *, 10> Candidates;
+      SmallVector<Instruction *, 10> AllCandidates;
+      SmallVector<Instruction *, 10> ColdCandidates;
       for (auto &I : BB.getInstList()) {
         const FunctionSamples *FS = nullptr;
         if ((isa<CallInst>(I) || isa<InvokeInst>(I)) &&
             !isa<IntrinsicInst>(I) && (FS = findCalleeFunctionSamples(I))) {
-          Candidates.push_back(&I);
+          AllCandidates.push_back(&I);
           if (FS->getEntrySamples() > 0)
             localNotInlinedCallSites.try_emplace(&I, FS);
           if (callsiteIsHot(FS, PSI))
             Hot = true;
+          else if (shouldInlineColdCallee(I))
+            ColdCandidates.push_back(&I);
         }
       }
       if (Hot) {
-        CIS.insert(CIS.begin(), Candidates.begin(), Candidates.end());
+        CIS.insert(CIS.begin(), AllCandidates.begin(), AllCandidates.end());
+      }
+      else {
+        CIS.insert(CIS.begin(), ColdCandidates.begin(), ColdCandidates.end());
       }
     }
     for (auto I : CIS) {
       Function *CalledFunction = CallSite(I).getCalledFunction();
       // Do not inline recursive calls.
       if (CalledFunction == &F)
         continue;
       if (CallSite(I).isIndirectCall()) {

llvm/test/Transforms/SampleProfile/Inputs/inline-cold.prof

+main:225715:0
+ 2.1: 5553
+ 3: 5391
+ 3.1: _Z3sumii:0
+  0: 0
+  1: 0
+  2: 0
 No newline at end of file

llvm/test/Transforms/SampleProfile/inline-cold.ll

+; Let sample profile loader replay inlining of small/cold functions
+
+; Make sure we don't inline the cold call sites by default
+; RUN: opt < %s -sample-profile -sample-profile-file=%S/Inputs/inline-cold.prof -S | FileCheck -check-prefix=NOTINLINE %s
+; RUN: opt < %s -passes=sample-profile -sample-profile-file=%S/Inputs/inline-cold.prof -S | FileCheck -check-prefix=NOTINLINE %s
+
+; Make sure we inline code call sites for size if requested
+; RUN: opt < %s -sample-profile -sample-profile-file=%S/Inputs/inline-cold.prof -sample-profile-inline-size -S | FileCheck -check-prefix=INLINE %s
+
+; Make sure we re-inline everything if requested 
+; RUN: opt < %s -passes=sample-profile -sample-profile-file=%S/Inputs/inline-cold.prof -sample-profile-inline-size -sample-profile-cold-inline-threshold=9999999 -S | FileCheck -check-prefix=INLINE %s
+
+; Make sure the separate size threshold for sample profile loader inlining works
+; RUN: opt < %s -passes=sample-profile -sample-profile-file=%S/Inputs/inline-cold.prof -sample-profile-inline-size -sample-profile-cold-inline-threshold=-500 -S | FileCheck -check-prefix=NOTINLINE %s
+
+@.str = private unnamed_addr constant [11 x i8] c"sum is %d\0A\00", align 1
+
+define i32 @_Z3sumii(i32 %x, i32 %y) !dbg !6 {
+entry:
+  %x.addr = alloca i32, align 4
+  %y.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  store i32 %y, i32* %y.addr, align 4
+  %tmp = load i32, i32* %x.addr, align 4, !dbg !8
+  %tmp1 = load i32, i32* %y.addr, align 4, !dbg !8
+  %add = add nsw i32 %tmp, %tmp1, !dbg !8
+  ret i32 %add, !dbg !8
+}
+
+define i32 @main() !dbg !9 {
+entry:
+  %retval = alloca i32, align 4
+  %s = alloca i32, align 4
+  %i = alloca i32, align 4
+  store i32 0, i32* %retval
+  store i32 0, i32* %i, align 4, !dbg !10
+  br label %while.cond, !dbg !11
+
+while.cond:                                       ; preds = %if.end, %entry
+  %tmp = load i32, i32* %i, align 4, !dbg !12
+  %inc = add nsw i32 %tmp, 1, !dbg !12
+  store i32 %inc, i32* %i, align 4, !dbg !12
+  %cmp = icmp slt i32 %tmp, 400000000, !dbg !12
+  br i1 %cmp, label %while.body, label %while.end, !dbg !12
+
+while.body:                                       ; preds = %while.cond
+  %tmp1 = load i32, i32* %i, align 4, !dbg !14
+  %cmp1 = icmp ne i32 %tmp1, 100, !dbg !14
+  br i1 %cmp1, label %if.then, label %if.else, !dbg !14
+
+if.then:                                          ; preds = %while.body
+  %tmp2 = load i32, i32* %i, align 4, !dbg !16
+  %tmp3 = load i32, i32* %s, align 4, !dbg !16
+  %call = call i32 @_Z3sumii(i32 %tmp2, i32 %tmp3), !dbg !16
+; INLINE-NOT: call i32 @_Z3sumii
+; NOTINLINE: call i32 @_Z3sumii
+  store i32 %call, i32* %s, align 4, !dbg !16
+  br label %if.end, !dbg !16
+
+if.else:                                          ; preds = %while.body
+  store i32 30, i32* %s, align 4, !dbg !18
+  br label %if.end
+
+if.end:                                           ; preds = %if.else, %if.then
+  br label %while.cond, !dbg !20
+
+while.end:                                        ; preds = %while.cond
+  %tmp4 = load i32, i32* %s, align 4, !dbg !22
+  %call2 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([11 x i8], [11 x i8]* @.str, i32 0, i32 0), i32 %tmp4), !dbg !22
+  ret i32 0, !dbg !23
+}
+
+declare i32 @printf(i8*, ...)
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!3, !4}
+!llvm.ident = !{!5}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, file: !1, producer: "clang version 3.5 ", isOptimized: false, runtimeVersion: 0, emissionKind: NoDebug, enums: !2, retainedTypes: !2, globals: !2, imports: !2)
+!1 = !DIFile(filename: "calls.cc", directory: ".")
+!2 = !{}
+!3 = !{i32 2, !"Dwarf Version", i32 4}
+!4 = !{i32 1, !"Debug Info Version", i32 3}
+!5 = !{!"clang version 3.5 "}
+!6 = distinct !DISubprogram(name: "sum", scope: !1, file: !1, line: 3, type: !7, scopeLine: 3, virtualIndex: 6, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !2)
+!7 = !DISubroutineType(types: !2)
+!8 = !DILocation(line: 4, scope: !6)
+!9 = distinct !DISubprogram(name: "main", scope: !1, file: !1, line: 7, type: !7, scopeLine: 7, virtualIndex: 6, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !2)
+!10 = !DILocation(line: 8, scope: !9)
+!11 = !DILocation(line: 9, scope: !9)
+!12 = !DILocation(line: 9, scope: !13)
+!13 = !DILexicalBlockFile(scope: !9, file: !1, discriminator: 2)
+!14 = !DILocation(line: 10, scope: !15)
+!15 = distinct !DILexicalBlock(scope: !9, file: !1, line: 10)
+!16 = !DILocation(line: 10, scope: !17)
+!17 = !DILexicalBlockFile(scope: !15, file: !1, discriminator: 2)
+!18 = !DILocation(line: 10, scope: !19)
+!19 = !DILexicalBlockFile(scope: !15, file: !1, discriminator: 4)
+!20 = !DILocation(line: 10, scope: !21)
+!21 = !DILexicalBlockFile(scope: !15, file: !1, discriminator: 6)
+!22 = !DILocation(line: 11, scope: !9)
+!23 = !DILocation(line: 12, scope: !9)