Merge remote-tracking branch 'upstream/release/12.x' into HEAD

Author: nikic

clang-tools-extra/docs/ReleaseNotes.rst

@@ -358,6 +358,16 @@ Changes in existing checks
 
   Added `std::basic_string_view` to default list of ``string``-like types.
 
+Deprecated checks
+^^^^^^^^^^^^^^^^^
+
+- The :doc:`readability-deleted-default
+  <clang-tidy/checks/readability-deleted-default>` check has been deprecated.
+  
+  The clang warning `Wdefaulted-function-deleted
+  <https://clang.llvm.org/docs/DiagnosticsReference.html#wdefaulted-function-deleted>`_
+  will diagnose the same issues and is enabled by default.
+
 Improvements to include-fixer
 -----------------------------
 

clang-tools-extra/docs/clang-tidy/checks/readability-deleted-default.rst

@@ -3,20 +3,6 @@
 readability-deleted-default
 ===========================
 
-Checks that constructors and assignment operators marked as ``= default`` are
-not actually deleted by the compiler.
-
-.. code-block:: c++
-
-  class Example {
-  public:
-    // This constructor is deleted because I is missing a default value.
-    Example() = default;
-    // This is fine.
-    Example(const Example& Other) = default;
-    // This operator is deleted because I cannot be assigned (it is const).
-    Example& operator=(const Example& Other) = default;
-
-  private:
-    const int I;
-  };
+This check has been deprecated prefer to make use of the `Wdefaulted-function-deleted
+<https://clang.llvm.org/docs/DiagnosticsReference.html#wdefaulted-function-deleted>`_
+flag.

clang/docs/ReleaseNotes.rst

@@ -73,6 +73,15 @@ New Compiler Flags
 
 - ...
 
+- AArch64 options ``-moutline-atomics``, ``-mno-outline-atomics`` to enable
+  and disable calls to helper functions implementing atomic operations. These
+  out-of-line helpers like '__aarch64_cas8_relax' will detect at runtime
+  AArch64 Large System Extensions (LSE) availability and either use their
+  atomic instructions, or falls back to LL/SC loop. These options do not apply
+  if the compilation target supports LSE. Atomic instructions are used directly
+  in that case. The option's behaviour mirrors GCC, the helpers are implemented
+  both in compiler-rt and libgcc.
+
 - -fpch-codegen and -fpch-debuginfo generate shared code and/or debuginfo
   for contents of a precompiled header in a separate object file. This object
   file needs to be linked in, but its contents do not need to be generated
@@ -368,7 +377,38 @@ libclang
 Static Analyzer
 ---------------
 
-- ...
+.. 3ff220de9009 [analyzer][StdLibraryFunctionsChecker] Add POSIX networking functions
+.. ...And a million other patches.
+- Improve the analyzer's understanding of several POSIX functions.
+
+.. https://reviews.llvm.org/D86533#2238207
+- Greatly improved the analyzer’s constraint solver by better understanding
+  when constraints are imposed on multiple symbolic values that are known to be
+  equal or known to be non-equal. It will now also efficiently reject impossible
+  if-branches between known comparison expressions. (Incorrectly stated as a
+  11.0.0 feature in the previous release notes)
+
+.. 820e8d8656ec [Analyzer][WebKit] UncountedLambdaCaptureChecker
+- New checker: :ref:`webkit.UncountedLambdaCapturesChecker<webkit-UncountedLambdaCapturesChecker>`
+  is a WebKit coding convention checker that flags raw pointers to
+  reference-counted objects captured by lambdas and suggests using intrusive
+  reference-counting smart pointers instead.
+
+.. 8a64689e264c [Analyzer][WebKit] UncountedLocalVarsChecker
+- New checker: :ref:`alpha.webkit.UncountedLocalVarsChecker<alpha-webkit-UncountedLocalVarsChecker>`
+  is a WebKit coding convention checker that intends to make sure that any
+  uncounted local variable is backed by a ref-counted object with lifetime that
+  is strictly larger than the scope of the uncounted local variable.
+
+.. i914f6c4ff8a4 [StaticAnalyzer] Support struct annotations in FuchsiaHandleChecker
+- ``fuchia.HandleChecker`` now recognizes handles in structs; All the handles
+  referenced by the structure (direct value or ptr) would be treated as
+  containing the release/use/acquire annotations directly.
+
+.. 8deaec122ec6 [analyzer] Update Fuchsia checker to catch releasing unowned handles.
+- Fuchsia checkers can detect the release of an unowned handle.
+
+- Numerous fixes and improvements to bug report generation.
 
 .. _release-notes-ubsan:
 

clang/docs/analyzer/checkers.rst

@@ -2538,6 +2538,8 @@ We also define a set of safe transformations which if passed a safe value as an
 - casts
 - unary operators like ``&`` or ``*``
 
+.. _alpha-webkit-UncountedLocalVarsChecker:
+
 alpha.webkit.UncountedLocalVarsChecker
 """"""""""""""""""""""""""""""""""""""
 The goal of this rule is to make sure that any uncounted local variable is backed by a ref-counted object with lifetime that is strictly larger than the scope of the uncounted local variable. To be on the safe side we require the scope of an uncounted variable to be embedded in the scope of ref-counted object that backs it.

clang/lib/Frontend/InitPreprocessor.cpp

@@ -565,7 +565,7 @@ static void InitializeCPlusPlusFeatureTestMacros(const LangOptions &LangOpts,
     Builder.defineMacro("__cpp_aggregate_bases", "201603L");
     Builder.defineMacro("__cpp_structured_bindings", "201606L");
     Builder.defineMacro("__cpp_nontype_template_args",
-                        LangOpts.CPlusPlus20 ? "201911L" : "201411L");
+                        "201411L"); // (not latest)
     Builder.defineMacro("__cpp_fold_expressions", "201603L");
     Builder.defineMacro("__cpp_guaranteed_copy_elision", "201606L");
     Builder.defineMacro("__cpp_nontype_template_parameter_auto", "201606L");

clang/test/Lexer/cxx-features.cpp

@@ -181,7 +181,8 @@
 #error "wrong value for __cpp_structured_bindings"
 #endif
 
-#if check(nontype_template_args, 0, 0, 0, 201411, 201911, 201911)
+#if check(nontype_template_args, 0, 0, 0, 201411, 201411, 201411)
+// FIXME: 201911 in C++20
 #error "wrong value for __cpp_nontype_template_args"
 #endif
 

compiler-rt/lib/builtins/CMakeLists.txt

@@ -515,14 +515,20 @@ set(aarch64_SOURCES
 set(OA_HELPERS_DIR "${CMAKE_CURRENT_BINARY_DIR}/outline_atomic_helpers.dir")
 file(MAKE_DIRECTORY "${OA_HELPERS_DIR}")
 
+if(CMAKE_HOST_UNIX)
+  set(COMPILER_RT_LINK_OR_COPY create_symlink)
+else()
+  set(COMPILER_RT_LINK_OR_COPY copy)
+endif()
+
 foreach(pat cas swp ldadd ldclr ldeor ldset)
   foreach(size 1 2 4 8 16)
     foreach(model 1 2 3 4)
       if(pat STREQUAL "cas" OR NOT size STREQUAL "16")
         set(helper_asm "${OA_HELPERS_DIR}/outline_atomic_${pat}${size}_${model}.S")
         add_custom_command(
           OUTPUT ${helper_asm}
-          COMMAND ${CMAKE_COMMAND} -E create_symlink "${CMAKE_CURRENT_SOURCE_DIR}/aarch64/lse.S" "${helper_asm}"
+          COMMAND ${CMAKE_COMMAND} -E ${COMPILER_RT_LINK_OR_COPY} "${CMAKE_CURRENT_SOURCE_DIR}/aarch64/lse.S" "${helper_asm}"
         )
         set_source_files_properties("${helper_asm}"
           PROPERTIES

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -6517,8 +6517,11 @@ static SDValue extractShiftForRotate(SelectionDAG &DAG, SDValue OppShift,
 // reduces to a rotate in direction shift2 by Pos or (equivalently) a rotate
 // in direction shift1 by Neg.  The range [0, EltSize) means that we only need
 // to consider shift amounts with defined behavior.
+//
+// The IsRotate flag should be set when the LHS of both shifts is the same.
+// Otherwise if matching a general funnel shift, it should be clear.
 static bool matchRotateSub(SDValue Pos, SDValue Neg, unsigned EltSize,
-                           SelectionDAG &DAG) {
+                           SelectionDAG &DAG, bool IsRotate) {
   // If EltSize is a power of 2 then:
   //
   //  (a) (Pos == 0 ? 0 : EltSize - Pos) == (EltSize - Pos) & (EltSize - 1)
@@ -6550,8 +6553,11 @@ static bool matchRotateSub(SDValue Pos, SDValue Neg, unsigned EltSize,
   // always invokes undefined behavior for 32-bit X.
   //
   // Below, Mask == EltSize - 1 when using [A] and is all-ones otherwise.
+  //
+  // NOTE: We can only do this when matching an AND and not a general
+  // funnel shift.
   unsigned MaskLoBits = 0;
-  if (Neg.getOpcode() == ISD::AND && isPowerOf2_64(EltSize)) {
+  if (IsRotate && Neg.getOpcode() == ISD::AND && isPowerOf2_64(EltSize)) {
     if (ConstantSDNode *NegC = isConstOrConstSplat(Neg.getOperand(1))) {
       KnownBits Known = DAG.computeKnownBits(Neg.getOperand(0));
       unsigned Bits = Log2_64(EltSize);
@@ -6641,7 +6647,8 @@ SDValue DAGCombiner::MatchRotatePosNeg(SDValue Shifted, SDValue Pos,
   //          (srl x, (*ext y))) ->
   //   (rotr x, y) or (rotl x, (sub 32, y))
   EVT VT = Shifted.getValueType();
-  if (matchRotateSub(InnerPos, InnerNeg, VT.getScalarSizeInBits(), DAG)) {
+  if (matchRotateSub(InnerPos, InnerNeg, VT.getScalarSizeInBits(), DAG,
+                     /*IsRotate*/ true)) {
     bool HasPos = TLI.isOperationLegalOrCustom(PosOpcode, VT);
     return DAG.getNode(HasPos ? PosOpcode : NegOpcode, DL, VT, Shifted,
                        HasPos ? Pos : Neg);
@@ -6670,7 +6677,7 @@ SDValue DAGCombiner::MatchFunnelPosNeg(SDValue N0, SDValue N1, SDValue Pos,
   // fold (or (shl x0, (*ext (sub 32, y))),
   //          (srl x1, (*ext y))) ->
   //   (fshr x0, x1, y) or (fshl x0, x1, (sub 32, y))
-  if (matchRotateSub(InnerPos, InnerNeg, EltBits, DAG)) {
+  if (matchRotateSub(InnerPos, InnerNeg, EltBits, DAG, /*IsRotate*/ N0 == N1)) {
     bool HasPos = TLI.isOperationLegalOrCustom(PosOpcode, VT);
     return DAG.getNode(HasPos ? PosOpcode : NegOpcode, DL, VT, N0, N1,
                        HasPos ? Pos : Neg);

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

@@ -5935,6 +5935,11 @@ SDValue TargetLowering::getNegatedExpression(SDValue Op, SelectionDAG &DAG,
 
   SDLoc DL(Op);
 
+  // Because getNegatedExpression can delete nodes we need a handle to keep
+  // temporary nodes alive in case the recursion manages to create an identical
+  // node.
+  std::list<HandleSDNode> Handles;
+
   switch (Opcode) {
   case ISD::ConstantFP: {
     // Don't invert constant FP values after legalization unless the target says
@@ -6003,11 +6008,18 @@ SDValue TargetLowering::getNegatedExpression(SDValue Op, SelectionDAG &DAG,
     NegatibleCost CostX = NegatibleCost::Expensive;
     SDValue NegX =
         getNegatedExpression(X, DAG, LegalOps, OptForSize, CostX, Depth);
+    // Prevent this node from being deleted by the next call.
+    if (NegX)
+      Handles.emplace_back(NegX);
+
     // fold (fneg (fadd X, Y)) -> (fsub (fneg Y), X)
     NegatibleCost CostY = NegatibleCost::Expensive;
     SDValue NegY =
         getNegatedExpression(Y, DAG, LegalOps, OptForSize, CostY, Depth);
 
+    // We're done with the handles.
+    Handles.clear();
+
     // Negate the X if its cost is less or equal than Y.
     if (NegX && (CostX <= CostY)) {
       Cost = CostX;
@@ -6052,11 +6064,18 @@ SDValue TargetLowering::getNegatedExpression(SDValue Op, SelectionDAG &DAG,
     NegatibleCost CostX = NegatibleCost::Expensive;
     SDValue NegX =
         getNegatedExpression(X, DAG, LegalOps, OptForSize, CostX, Depth);
+    // Prevent this node from being deleted by the next call.
+    if (NegX)
+      Handles.emplace_back(NegX);
+
     // fold (fneg (fmul X, Y)) -> (fmul X, (fneg Y))
     NegatibleCost CostY = NegatibleCost::Expensive;
     SDValue NegY =
         getNegatedExpression(Y, DAG, LegalOps, OptForSize, CostY, Depth);
 
+    // We're done with the handles.
+    Handles.clear();
+
     // Negate the X if its cost is less or equal than Y.
     if (NegX && (CostX <= CostY)) {
       Cost = CostX;
@@ -6094,15 +6113,25 @@ SDValue TargetLowering::getNegatedExpression(SDValue Op, SelectionDAG &DAG,
     if (!NegZ)
       break;
 
+    // Prevent this node from being deleted by the next two calls.
+    Handles.emplace_back(NegZ);
+
     // fold (fneg (fma X, Y, Z)) -> (fma (fneg X), Y, (fneg Z))
     NegatibleCost CostX = NegatibleCost::Expensive;
     SDValue NegX =
         getNegatedExpression(X, DAG, LegalOps, OptForSize, CostX, Depth);
+    // Prevent this node from being deleted by the next call.
+    if (NegX)
+      Handles.emplace_back(NegX);
+
     // fold (fneg (fma X, Y, Z)) -> (fma X, (fneg Y), (fneg Z))
     NegatibleCost CostY = NegatibleCost::Expensive;
     SDValue NegY =
         getNegatedExpression(Y, DAG, LegalOps, OptForSize, CostY, Depth);
 
+    // We're done with the handles.
+    Handles.clear();
+
     // Negate the X if its cost is less or equal than Y.
     if (NegX && (CostX <= CostY)) {
       Cost = std::min(CostX, CostZ);

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -1017,11 +1017,12 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
     // Vector reductions
     for (MVT VT : { MVT::v4f16, MVT::v2f32,
                     MVT::v8f16, MVT::v4f32, MVT::v2f64 }) {
-      setOperationAction(ISD::VECREDUCE_FMAX, VT, Custom);
-      setOperationAction(ISD::VECREDUCE_FMIN, VT, Custom);
+      if (VT.getVectorElementType() != MVT::f16 || Subtarget->hasFullFP16()) {
+        setOperationAction(ISD::VECREDUCE_FMAX, VT, Custom);
+        setOperationAction(ISD::VECREDUCE_FMIN, VT, Custom);
 
-      if (VT.getVectorElementType() != MVT::f16 || Subtarget->hasFullFP16())
         setOperationAction(ISD::VECREDUCE_FADD, VT, Legal);
+      }
     }
     for (MVT VT : { MVT::v8i8, MVT::v4i16, MVT::v2i32,
                     MVT::v16i8, MVT::v8i16, MVT::v4i32 }) {

llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp

@@ -345,10 +345,14 @@ Value *InstCombinerImpl::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
         return false;
 
       // Get the constant out of the ICmp, if there is one.
+      // Only try this when exactly 1 operand is a constant (if both operands
+      // are constant, the icmp should eventually simplify). Otherwise, we may
+      // invert the transform that reduces set bits and infinite-loop.
+      Value *X;
       const APInt *CmpC;
       ICmpInst::Predicate Pred;
-      if (!match(I->getOperand(0), m_c_ICmp(Pred, m_APInt(CmpC), m_Value())) ||
-          CmpC->getBitWidth() != SelC->getBitWidth())
+      if (!match(I->getOperand(0), m_ICmp(Pred, m_Value(X), m_APInt(CmpC))) ||
+          isa<Constant>(X) || CmpC->getBitWidth() != SelC->getBitWidth())
         return ShrinkDemandedConstant(I, OpNo, DemandedMask);
 
       // If the constant is already the same as the ICmp, leave it as-is.

llvm/lib/Transforms/Utils/Local.cpp

@@ -918,17 +918,39 @@ static void gatherIncomingValuesToPhi(PHINode *PN,
 /// \param IncomingValues A map from block to value.
 static void replaceUndefValuesInPhi(PHINode *PN,
                                     const IncomingValueMap &IncomingValues) {
+  SmallVector<unsigned> TrueUndefOps;
   for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
     Value *V = PN->getIncomingValue(i);
 
     if (!isa<UndefValue>(V)) continue;
 
     BasicBlock *BB = PN->getIncomingBlock(i);
     IncomingValueMap::const_iterator It = IncomingValues.find(BB);
-    if (It == IncomingValues.end()) continue;
 
+    // Keep track of undef/poison incoming values. Those must match, so we fix
+    // them up below if needed.
+    // Note: this is conservatively correct, but we could try harder and group
+    // the undef values per incoming basic block.
+    if (It == IncomingValues.end()) {
+      TrueUndefOps.push_back(i);
+      continue;
+    }
+
+    // There is a defined value for this incoming block, so map this undef
+    // incoming value to the defined value.
     PN->setIncomingValue(i, It->second);
   }
+
+  // If there are both undef and poison values incoming, then convert those
+  // values to undef. It is invalid to have different values for the same
+  // incoming block.
+  unsigned PoisonCount = count_if(TrueUndefOps, [&](unsigned i) {
+    return isa<PoisonValue>(PN->getIncomingValue(i));
+  });
+  if (PoisonCount != 0 && PoisonCount != TrueUndefOps.size()) {
+    for (unsigned i : TrueUndefOps)
+      PN->setIncomingValue(i, UndefValue::get(PN->getType()));
+  }
 }
 
 /// Replace a value flowing from a block to a phi with

llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h

@@ -142,6 +142,10 @@ class VPBuilder {
     return createInstruction(Instruction::BinaryOps::Or, {LHS, RHS});
   }
 
+  VPValue *createSelect(VPValue *Cond, VPValue *TrueVal, VPValue *FalseVal) {
+    return createNaryOp(Instruction::Select, {Cond, TrueVal, FalseVal});
+  }
+
   //===--------------------------------------------------------------------===//
   // RAII helpers.
   //===--------------------------------------------------------------------===//

llvm/lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -8195,8 +8195,15 @@ VPValue *VPRecipeBuilder::createEdgeMask(BasicBlock *Src, BasicBlock *Dst,
   if (BI->getSuccessor(0) != Dst)
     EdgeMask = Builder.createNot(EdgeMask);
 
-  if (SrcMask) // Otherwise block in-mask is all-one, no need to AND.
-    EdgeMask = Builder.createAnd(EdgeMask, SrcMask);
+  if (SrcMask) { // Otherwise block in-mask is all-one, no need to AND.
+    // The condition is 'SrcMask && EdgeMask', which is equivalent to
+    // 'select i1 SrcMask, i1 EdgeMask, i1 false'.
+    // The select version does not introduce new UB if SrcMask is false and
+    // EdgeMask is poison. Using 'and' here introduces undefined behavior.
+    VPValue *False = Plan->getOrAddVPValue(
+        ConstantInt::getFalse(BI->getCondition()->getType()));
+    EdgeMask = Builder.createSelect(SrcMask, EdgeMask, False);
+  }
 
   return EdgeMaskCache[Edge] = EdgeMask;
 }