merge main into amd-staging

Author: z1.cciauto

clang/include/clang/AST/DeclTemplate.h

@@ -781,16 +781,15 @@ class RedeclarableTemplateDecl : public TemplateDecl,
   bool loadLazySpecializationsImpl(llvm::ArrayRef<TemplateArgument> Args,
                                    TemplateParameterList *TPL = nullptr) const;
 
-  template <class EntryType, typename ...ProfileArguments>
-  typename SpecEntryTraits<EntryType>::DeclType*
+  template <class EntryType, typename... ProfileArguments>
+  typename SpecEntryTraits<EntryType>::DeclType *
   findSpecializationImpl(llvm::FoldingSetVector<EntryType> &Specs,
-                         void *&InsertPos, ProfileArguments &&...ProfileArgs);
+                         void *&InsertPos, ProfileArguments... ProfileArgs);
 
   template <class EntryType, typename... ProfileArguments>
   typename SpecEntryTraits<EntryType>::DeclType *
   findSpecializationLocally(llvm::FoldingSetVector<EntryType> &Specs,
-                            void *&InsertPos,
-                            ProfileArguments &&...ProfileArgs);
+                            void *&InsertPos, ProfileArguments... ProfileArgs);
 
   template <class Derived, class EntryType>
   void addSpecializationImpl(llvm::FoldingSetVector<EntryType> &Specs,

clang/include/clang/Basic/SourceManager.h

@@ -1529,6 +1529,15 @@ class SourceManager : public RefCountedBase<SourceManager> {
     return Filename == "<scratch space>";
   }
 
+  /// Returns whether \p Loc is located in a built-in or command line source.
+  bool isInPredefinedFile(SourceLocation Loc) const {
+    PresumedLoc Presumed = getPresumedLoc(Loc);
+    if (Presumed.isInvalid())
+      return false;
+    StringRef Filename(Presumed.getFilename());
+    return Filename == "<built-in>" || Filename == "<command line>";
+  }
+
   /// Returns if a SourceLocation is in a system header.
   bool isInSystemHeader(SourceLocation Loc) const {
     if (Loc.isInvalid())

clang/include/clang/StaticAnalyzer/Core/AnalyzerOptions.def

@@ -414,6 +414,19 @@ ANALYZER_OPTION(
     "any target.",
     true)
 
+ANALYZER_OPTION(
+    bool, InlineFunctionsWithAmbiguousLoops, "inline-functions-with-ambiguous-loops",
+    "If disabled (the default), the analyzer puts functions on a \"do not "
+    "inline this\" list if it finds an execution path within that function "
+    "that may potentially perform 'analyzer-max-loop' (= 4 by default) "
+    "iterations in a loop. (Note that functions that _definitely_ reach the "
+    "loop limit on some execution path are currently marked as \"do not "
+    "inline\" even if this option is enabled.) Enabling this option "
+    "eliminates this (somewhat arbitrary) restriction from the analysis "
+    "scope, which increases the analysis runtime (on average by ~10%, but "
+    "a few translation units may see much larger slowdowns).",
+    false)
+
 //===----------------------------------------------------------------------===//
 // Unsigned analyzer options.
 //===----------------------------------------------------------------------===//

clang/include/clang/StaticAnalyzer/Core/PathSensitive/FunctionSummary.h

@@ -81,10 +81,6 @@ class FunctionSummariesTy {
     I->second.MayInline = 0;
   }
 
-  void markReachedMaxBlockCount(const Decl *D) {
-    markShouldNotInline(D);
-  }
-
   std::optional<bool> mayInline(const Decl *D) {
     MapTy::const_iterator I = Map.find(D);
     if (I != Map.end() && I->second.InlineChecked)

clang/lib/AST/DeclTemplate.cpp

@@ -382,12 +382,11 @@ template <class EntryType, typename... ProfileArguments>
 typename RedeclarableTemplateDecl::SpecEntryTraits<EntryType>::DeclType *
 RedeclarableTemplateDecl::findSpecializationLocally(
     llvm::FoldingSetVector<EntryType> &Specs, void *&InsertPos,
-    ProfileArguments &&...ProfileArgs) {
+    ProfileArguments... ProfileArgs) {
   using SETraits = RedeclarableTemplateDecl::SpecEntryTraits<EntryType>;
 
   llvm::FoldingSetNodeID ID;
-  EntryType::Profile(ID, std::forward<ProfileArguments>(ProfileArgs)...,
-                     getASTContext());
+  EntryType::Profile(ID, ProfileArgs..., getASTContext());
   EntryType *Entry = Specs.FindNodeOrInsertPos(ID, InsertPos);
   return Entry ? SETraits::getDecl(Entry)->getMostRecentDecl() : nullptr;
 }
@@ -396,18 +395,15 @@ template <class EntryType, typename... ProfileArguments>
 typename RedeclarableTemplateDecl::SpecEntryTraits<EntryType>::DeclType *
 RedeclarableTemplateDecl::findSpecializationImpl(
     llvm::FoldingSetVector<EntryType> &Specs, void *&InsertPos,
-    ProfileArguments &&...ProfileArgs) {
+    ProfileArguments... ProfileArgs) {
 
-  if (auto *Found = findSpecializationLocally(
-          Specs, InsertPos, std::forward<ProfileArguments>(ProfileArgs)...))
+  if (auto *Found = findSpecializationLocally(Specs, InsertPos, ProfileArgs...))
     return Found;
 
-  if (!loadLazySpecializationsImpl(
-          std::forward<ProfileArguments>(ProfileArgs)...))
+  if (!loadLazySpecializationsImpl(ProfileArgs...))
     return nullptr;
 
-  return findSpecializationLocally(
-      Specs, InsertPos, std::forward<ProfileArguments>(ProfileArgs)...);
+  return findSpecializationLocally(Specs, InsertPos, ProfileArgs...);
 }
 
 template<class Derived, class EntryType>

clang/lib/Driver/Driver.cpp

@@ -5646,19 +5646,20 @@ class ToolSelector final {
     if (!BJ || !CJ)
       return nullptr;
 
+    auto HasBitcodeInput = [](const JobActionInfo &AI) {
+      for (auto &Input : AI.JA->getInputs())
+        if (!types::isLLVMIR(Input->getType()))
+          return false;
+      return true;
+    };
+
     // Check if the initial input (to the compile job or its predessor if one
     // exists) is LLVM bitcode. In that case, no preprocessor step is required
     // and we can still collapse the compile and backend jobs when we have
     // -save-temps. I.e. there is no need for a separate compile job just to
     // emit unoptimized bitcode.
-    bool InputIsBitcode = true;
-    for (size_t i = 1; i < ActionInfo.size(); i++)
-      if (ActionInfo[i].JA->getType() != types::TY_LLVM_BC &&
-          ActionInfo[i].JA->getType() != types::TY_LTO_BC) {
-        InputIsBitcode = false;
-        break;
-      }
-    if (!InputIsBitcode && !canCollapsePreprocessorAction())
+    bool InputIsBitcode = all_of(ActionInfo, HasBitcodeInput);
+    if (SaveTemps && !InputIsBitcode)
       return nullptr;
 
     // Get compiler tool.
@@ -5672,7 +5673,7 @@ class ToolSelector final {
     if (!T->hasIntegratedBackend() && !(OutputIsLLVM && T->canEmitIR()))
       return nullptr;
 
-    if (T->canEmitIR() && ((SaveTemps && !InputIsBitcode) || EmbedBitcode))
+    if (T->canEmitIR() && EmbedBitcode)
       return nullptr;
 
     Inputs = CJ->getInputs();

clang/lib/Driver/ToolChains/Clang.cpp

@@ -549,7 +549,7 @@ static void addDashXForInput(const ArgList &Args, const InputInfo &Input,
 
   CmdArgs.push_back("-x");
   if (Args.hasArg(options::OPT_rewrite_objc))
-    CmdArgs.push_back(types::getTypeName(types::TY_PP_ObjCXX));
+    CmdArgs.push_back(types::getTypeName(types::TY_ObjCXX));
   else {
     // Map the driver type to the frontend type. This is mostly an identity
     // mapping, except that the distinction between module interface units

clang/lib/ExtractAPI/ExtractAPIConsumer.cpp

@@ -305,8 +305,7 @@ class MacroCallback : public PPCallbacks {
 
       auto DefLoc = MI->getDefinitionLoc();
 
-      if (SM.isWrittenInBuiltinFile(DefLoc) ||
-          SM.isWrittenInCommandLineFile(DefLoc))
+      if (SM.isInPredefinedFile(DefLoc))
         continue;
 
       auto AssociatedModuleMacros = MD.getModuleMacros();

clang/lib/Frontend/PrintPreprocessedOutput.cpp

@@ -569,8 +569,7 @@ void PrintPPOutputPPCallbacks::MacroDefined(const Token &MacroNameTok,
   SourceLocation DefLoc = MI->getDefinitionLoc();
   if (DirectivesOnly && !MI->isUsed()) {
     SourceManager &SM = PP.getSourceManager();
-    if (SM.isWrittenInBuiltinFile(DefLoc) ||
-        SM.isWrittenInCommandLineFile(DefLoc))
+    if (SM.isInPredefinedFile(DefLoc))
       return;
   }
   MoveToLine(DefLoc, /*RequireStartOfLine=*/true);

clang/lib/Lex/PPDirectives.cpp

@@ -374,9 +374,8 @@ bool Preprocessor::CheckMacroName(Token &MacroNameTok, MacroUse isDefineUndef,
   // Macro names with reserved identifiers are accepted if built-in or passed
   // through the command line (the later may be present if -dD was used to
   // generate the preprocessed file).
-  bool IsBuiltinOrCmd = SourceMgr.isWrittenInBuiltinFile(MacroNameLoc) ||
-                        SourceMgr.isWrittenInCommandLineFile(MacroNameLoc);
-  if (!IsBuiltinOrCmd && !SourceMgr.isInSystemHeader(MacroNameLoc)) {
+  if (!SourceMgr.isInPredefinedFile(MacroNameLoc) &&
+      !SourceMgr.isInSystemHeader(MacroNameLoc)) {
     MacroDiag D = MD_NoWarn;
     if (isDefineUndef == MU_Define) {
       D = shouldWarnOnMacroDef(*this, II);
@@ -1706,8 +1705,7 @@ void Preprocessor::HandleDigitDirective(Token &DigitTok) {
     // If a filename was present, read any flags that are present.
     if (ReadLineMarkerFlags(IsFileEntry, IsFileExit, FileKind, *this))
       return;
-    if (!SourceMgr.isWrittenInBuiltinFile(DigitTok.getLocation()) &&
-        !SourceMgr.isWrittenInCommandLineFile(DigitTok.getLocation()))
+    if (!SourceMgr.isInPredefinedFile(DigitTok.getLocation()))
       Diag(StrTok, diag::ext_pp_gnu_line_directive);
 
     // Exiting to an empty string means pop to the including file, so leave

clang/lib/StaticAnalyzer/Core/ExprEngine.cpp

@@ -2523,6 +2523,20 @@ bool ExprEngine::replayWithoutInlining(ExplodedNode *N,
   return true;
 }
 
+/// Return the innermost location context which is inlined at `Node`, unless
+/// it's the top-level (entry point) location context.
+static const LocationContext *getInlinedLocationContext(ExplodedNode *Node,
+                                                        ExplodedGraph &G) {
+  const LocationContext *CalleeLC = Node->getLocation().getLocationContext();
+  const LocationContext *RootLC =
+      (*G.roots_begin())->getLocation().getLocationContext();
+
+  if (CalleeLC->getStackFrame() == RootLC->getStackFrame())
+    return nullptr;
+
+  return CalleeLC;
+}
+
 /// Block entrance.  (Update counters).
 void ExprEngine::processCFGBlockEntrance(const BlockEdge &L,
                                          NodeBuilderWithSinks &nodeBuilder,
@@ -2570,21 +2584,24 @@ void ExprEngine::processCFGBlockEntrance(const BlockEdge &L,
     const ExplodedNode *Sink =
                    nodeBuilder.generateSink(Pred->getState(), Pred, &tag);
 
-    // Check if we stopped at the top level function or not.
-    // Root node should have the location context of the top most function.
-    const LocationContext *CalleeLC = Pred->getLocation().getLocationContext();
-    const LocationContext *CalleeSF = CalleeLC->getStackFrame();
-    const LocationContext *RootLC =
-                        (*G.roots_begin())->getLocation().getLocationContext();
-    if (RootLC->getStackFrame() != CalleeSF) {
-      Engine.FunctionSummaries->markReachedMaxBlockCount(CalleeSF->getDecl());
+    if (const LocationContext *LC = getInlinedLocationContext(Pred, G)) {
+      // FIXME: This will unconditionally prevent inlining this function (even
+      // from other entry points), which is not a reasonable heuristic: even if
+      // we reached max block count on this particular execution path, there
+      // may be other execution paths (especially with other parametrizations)
+      // where the analyzer can reach the end of the function (so there is no
+      // natural reason to avoid inlining it). However, disabling this would
+      // significantly increase the analysis time (because more entry points
+      // would exhaust their allocated budget), so it must be compensated by a
+      // different (more reasonable) reduction of analysis scope.
+      Engine.FunctionSummaries->markShouldNotInline(
+          LC->getStackFrame()->getDecl());
 
       // Re-run the call evaluation without inlining it, by storing the
       // no-inlining policy in the state and enqueuing the new work item on
       // the list. Replay should almost never fail. Use the stats to catch it
       // if it does.
-      if ((!AMgr.options.NoRetryExhausted &&
-           replayWithoutInlining(Pred, CalleeLC)))
+      if ((!AMgr.options.NoRetryExhausted && replayWithoutInlining(Pred, LC)))
         return;
       NumMaxBlockCountReachedInInlined++;
     } else
@@ -2856,8 +2873,29 @@ void ExprEngine::processBranch(
       // conflicts with the widen-loop analysis option (which is off by
       // default). If we intend to support and stabilize the loop widening,
       // we must ensure that it 'plays nicely' with this logic.
-      if (!SkipTrueBranch || AMgr.options.ShouldWidenLoops)
+      if (!SkipTrueBranch || AMgr.options.ShouldWidenLoops) {
         Builder.generateNode(StTrue, true, PredN);
+      } else if (!AMgr.options.InlineFunctionsWithAmbiguousLoops) {
+        // FIXME: There is an ancient and arbitrary heuristic in
+        // `ExprEngine::processCFGBlockEntrance` which prevents all further
+        // inlining of a function if it finds an execution path within that
+        // function which reaches the `MaxBlockVisitOnPath` limit (a/k/a
+        // `analyzer-max-loop`, by default four iterations in a loop). Adding
+        // this "don't assume third iteration" logic significantly increased
+        // the analysis runtime on some inputs because less functions were
+        // arbitrarily excluded from being inlined, so more entry points used
+        // up their full allocated budget. As a hacky compensation for this,
+        // here we apply the "should not inline" mark in cases when the loop
+        // could potentially reach the `MaxBlockVisitOnPath` limit without the
+        // "don't assume third iteration" logic. This slightly overcompensates
+        // (activates if the third iteration can be entered, and will not
+        // recognize cases where the fourth iteration would't be completed), but
+        // should be good enough for practical purposes.
+        if (const LocationContext *LC = getInlinedLocationContext(Pred, G)) {
+          Engine.FunctionSummaries->markShouldNotInline(
+              LC->getStackFrame()->getDecl());
+        }
+      }
     }
 
     if (StFalse) {

clang/test/Analysis/analyzer-config.c

@@ -89,6 +89,7 @@
 // CHECK-NEXT: graph-trim-interval = 1000
 // CHECK-NEXT: ignore-bison-generated-files = true
 // CHECK-NEXT: ignore-flex-generated-files = true
+// CHECK-NEXT: inline-functions-with-ambiguous-loops = false
 // CHECK-NEXT: inline-lambdas = true
 // CHECK-NEXT: ipa = dynamic-bifurcate
 // CHECK-NEXT: ipa-always-inline-size = 3