[analyzer] Implement "do not inline large functions many times"
performance heuristic
After inlining a function with more than 13 basic blocks 32 times, we
are not going to inline it anymore. The idea is that inlining large
functions leads to drastic performance implications. Since the function
has already been inlined, we know that we've analyzed it in many
contexts.
The following metrics are used:
- Large function is a function with more than 13 basic blocks (we
should switch to another metric, like cyclomatic complexity)
- We consider that we've inlined a function many times if it's been
inlined 32 times. This number is configurable with -analyzer-config
max-times-inline-large=xx
This heuristic addresses a performance regression introduced with
inlining on one benchmark. The analyzer on this benchmark became 60
times slower with inlining turned on. The heuristic allows us to analyze
it in 24% of the time. The performance improvements on the other
benchmarks I've tested with are much lower - under 10%, which is
expected.
git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@170361 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/include/clang/StaticAnalyzer/Core/AnalyzerOptions.h b/include/clang/StaticAnalyzer/Core/AnalyzerOptions.h
index 9cbce17..02bb036 100644
--- a/include/clang/StaticAnalyzer/Core/AnalyzerOptions.h
+++ b/include/clang/StaticAnalyzer/Core/AnalyzerOptions.h
@@ -193,6 +193,9 @@
/// \sa getGraphTrimInterval
llvm::Optional<unsigned> GraphTrimInterval;
+ /// \sa getMaxTimesInlineLarge
+ llvm::Optional<unsigned> MaxTimesInlineLarge;
+
/// Interprets an option's string value as a boolean.
///
/// Accepts the strings "true" and "false".
@@ -276,6 +279,11 @@
/// node reclamation, set the option to "0".
unsigned getGraphTrimInterval();
+ /// Returns the maximum times a large function could be inlined.
+ ///
+ /// This is controlled by the 'max-times-inline-large' config option.
+ unsigned getMaxTimesInlineLarge();
+
public:
AnalyzerOptions() : CXXMemberInliningMode() {
AnalysisStoreOpt = RegionStoreModel;
diff --git a/include/clang/StaticAnalyzer/Core/PathSensitive/FunctionSummary.h b/include/clang/StaticAnalyzer/Core/PathSensitive/FunctionSummary.h
index 9e2505f..546cec5 100644
--- a/include/clang/StaticAnalyzer/Core/PathSensitive/FunctionSummary.h
+++ b/include/clang/StaticAnalyzer/Core/PathSensitive/FunctionSummary.h
@@ -37,10 +37,14 @@
/// Marks the IDs of the basic blocks visited during the analyzes.
llvm::BitVector VisitedBasicBlocks;
+ /// The number of times the function has been inlined.
+ unsigned TimesInlined;
+
FunctionSummary() :
MayReachMaxBlockCount(false),
TotalBasicBlocks(0),
- VisitedBasicBlocks(0) {}
+ VisitedBasicBlocks(0),
+ TimesInlined(0) {}
};
typedef llvm::DenseMap<const Decl*, FunctionSummary*> MapTy;
@@ -84,11 +88,23 @@
unsigned getNumVisitedBasicBlocks(const Decl* D) {
MapTy::const_iterator I = Map.find(D);
- if (I != Map.end())
- return I->second->VisitedBasicBlocks.count();
+ if (I != Map.end())
+ return I->second->VisitedBasicBlocks.count();
return 0;
}
+ unsigned getNumTimesInlined(const Decl* D) {
+ MapTy::const_iterator I = Map.find(D);
+ if (I != Map.end())
+ return I->second->TimesInlined;
+ return 0;
+ }
+
+ void bumpNumTimesInlined(const Decl* D) {
+ MapTy::iterator I = findOrInsertSummary(D);
+ I->second->TimesInlined++;
+ }
+
/// Get the percentage of the reachable blocks.
unsigned getPercentBlocksReachable(const Decl *D) {
MapTy::const_iterator I = Map.find(D);
diff --git a/lib/StaticAnalyzer/Core/AnalyzerOptions.cpp b/lib/StaticAnalyzer/Core/AnalyzerOptions.cpp
index 8ecbac0..b993804 100644
--- a/lib/StaticAnalyzer/Core/AnalyzerOptions.cpp
+++ b/lib/StaticAnalyzer/Core/AnalyzerOptions.cpp
@@ -133,6 +133,12 @@
return GraphTrimInterval.getValue();
}
+unsigned AnalyzerOptions::getMaxTimesInlineLarge() {
+ if (!MaxTimesInlineLarge.hasValue())
+ MaxTimesInlineLarge = getOptionAsInteger("max-times-inline-large", 32);
+ return MaxTimesInlineLarge.getValue();
+}
+
bool AnalyzerOptions::shouldSynthesizeBodies() {
return getBooleanOption("faux-bodies", true);
}
diff --git a/lib/StaticAnalyzer/Core/ExprEngineCallAndReturn.cpp b/lib/StaticAnalyzer/Core/ExprEngineCallAndReturn.cpp
index d654bb5..7c1c26e 100644
--- a/lib/StaticAnalyzer/Core/ExprEngineCallAndReturn.cpp
+++ b/lib/StaticAnalyzer/Core/ExprEngineCallAndReturn.cpp
@@ -33,6 +33,9 @@
STATISTIC(NumInlinedCalls,
"The # of times we inlined a call");
+STATISTIC(NumReachedInlineCountMax,
+ "The # of times we reached inline count maximum");
+
void ExprEngine::processCallEnter(CallEnter CE, ExplodedNode *Pred) {
// Get the entry block in the CFG of the callee.
const StackFrameContext *calleeCtx = CE.getCalleeContext();
@@ -415,12 +418,12 @@
if (getContext().getLangOpts().CPlusPlus) {
if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) {
// Conditionally allow the inlining of template functions.
- if (!getAnalysisManager().options.mayInlineTemplateFunctions())
+ if (!AMgr.options.mayInlineTemplateFunctions())
if (FD->getTemplatedKind() != FunctionDecl::TK_NonTemplate)
return false;
// Conditionally allow the inlining of C++ standard library functions.
- if (!getAnalysisManager().options.mayInlineCXXStandardLibrary())
+ if (!AMgr.options.mayInlineCXXStandardLibrary())
if (getContext().getSourceManager().isInSystemHeader(FD->getLocation()))
if (IsInStdNamespace(FD))
return false;
@@ -432,6 +435,14 @@
if (!CalleeADC->getAnalysis<RelaxedLiveVariables>())
return false;
+ if (Engine.FunctionSummaries->getNumTimesInlined(D) >
+ AMgr.options.getMaxTimesInlineLarge() &&
+ CalleeCFG->getNumBlockIDs() > 13) {
+ NumReachedInlineCountMax++;
+ return false;
+ }
+ Engine.FunctionSummaries->bumpNumTimesInlined(D);
+
return true;
}
diff --git a/test/Analysis/analyzer-config.c b/test/Analysis/analyzer-config.c
index 990f578..d156a6e 100644
--- a/test/Analysis/analyzer-config.c
+++ b/test/Analysis/analyzer-config.c
@@ -9,5 +9,6 @@
// CHECK-NEXT: faux-bodies = true
// CHECK-NEXT: graph-trim-interval = 1000
// CHECK-NEXT: ipa-always-inline-size = 3
+// CHECK-NEXT: max-times-inline-large = 32
// CHECK-NEXT: [stats]
-// CHECK-NEXT: num-entries = 4
+// CHECK-NEXT: num-entries = 5
diff --git a/test/Analysis/analyzer-config.cpp b/test/Analysis/analyzer-config.cpp
index fb14266..b3a8b34 100644
--- a/test/Analysis/analyzer-config.cpp
+++ b/test/Analysis/analyzer-config.cpp
@@ -18,5 +18,6 @@
// CHECK-NEXT: faux-bodies = true
// CHECK-NEXT: graph-trim-interval = 1000
// CHECK-NEXT: ipa-always-inline-size = 3
+// CHECK-NEXT: max-times-inline-large = 32
// CHECK-NEXT: [stats]
-// CHECK-NEXT: num-entries = 7
+// CHECK-NEXT: num-entries = 8
