[pull] main from llvm:main

.github/CODEOWNERS

@@ -53,6 +53,13 @@
 /mlir/include/mlir/Interfaces/DestinationStyleOpInterface.* @matthias-springer
 /mlir/lib/Interfaces/DestinationStyleOpInterface.* @matthias-springer
 
+# AMDGPU and ROCDL dialects in MLIR.
+/mlir/include/mlir/Dialect/AMDGPU @krzysz00 @kuhar
+/mlir/lib/Dialect/AMDGPU @krzysz00 @kuhar
+/mlir/lib/Conversion/*AMDGPU* @krzysz00 @kuhar
+/mlir/lib/Conversion/*ToROCDL @krzysz00 @kuhar
+/mlir/include/mlir/Dialect/LLVMIR/ROCDL* @krzysz00 @kuhar
+
 # Bufferization Dialect in MLIR.
 /mlir/include/mlir/Dialect/Bufferization @matthias-springer
 /mlir/lib/Dialect/Bufferization @matthias-springer

.github/workflows/pr-code-lint.yml

@@ -20,7 +20,7 @@ jobs:
       run:
         shell: bash
     container:
-      image: 'ghcr.io/llvm/ci-ubuntu-24.04:latest'
+      image: 'ghcr.io/llvm/ci-ubuntu-24.04-lint'
     timeout-minutes: 60
     concurrency:
       group: ${{ github.workflow }}-${{ github.ref }}
@@ -31,6 +31,11 @@ jobs:
         with:
           fetch-depth: 2
 
+      # FIXME: same as in ".github/workflows/pr-code-format.yml"
+      - name: Set Safe Directory
+        run: |
+          chown -R root $(pwd)
+
       - name: Get changed files
         id: changed-files
         uses: tj-actions/changed-files@ed68ef82c095e0d48ec87eccea555d944a631a4c # v46.0.5
@@ -46,22 +51,6 @@ jobs:
         run: |
           echo "Changed files:"
           echo "$CHANGED_FILES"
-
-      # The clang tidy version should always be upgraded to the first version
-      # of a release cycle (x.1.0) or the last version of a release cycle, or
-      # if there have been relevant clang-format backports.
-      - name: Install clang-tidy
-        uses: aminya/setup-cpp@a276e6e3d1db9160db5edc458e99a30d3b109949 # v1.7.1
-        with:
-          clang-tidy: 21.1.0
-
-      - name: Setup Python env
-        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
-        with:
-          python-version: '3.13'
-
-      - name: Install Python dependencies
-        run: python3 -m pip install -r llvm/utils/git/requirements_linting.txt
 
       # TODO: create special mapping for 'codegen' targets, for now build predefined set
       # TODO: add entrypoint in 'compute_projects.py' that only adds a project and its direct dependencies

clang-tools-extra/clangd/unittests/lit.cfg.py

@@ -19,12 +19,12 @@
 
 if platform.system() == "Darwin":
     shlibpath_var = "DYLD_LIBRARY_PATH"
-elif platform.system() == "Windows":
+elif platform.system() == "Windows" or sys.platform == "cygwin":
     shlibpath_var = "PATH"
 else:
     shlibpath_var = "LD_LIBRARY_PATH"
 config.environment[shlibpath_var] = os.path.pathsep.join(
-    ("@SHLIBDIR@", "@LLVM_LIBS_DIR@", config.environment.get(shlibpath_var, ""))
+    (config.shlibdir, config.llvm_libs_dir, config.environment.get(shlibpath_var, ""))
 )
 
 # It is not realistically possible to account for all options that could

clang-tools-extra/include-cleaner/test/Unit/lit.cfg.py

@@ -11,12 +11,12 @@
 
 if platform.system() == "Darwin":
     shlibpath_var = "DYLD_LIBRARY_PATH"
-elif platform.system() == "Windows":
+elif platform.system() == "Windows" or sys.platform == "cygwin":
     shlibpath_var = "PATH"
 else:
     shlibpath_var = "LD_LIBRARY_PATH"
 config.environment[shlibpath_var] = os.path.pathsep.join(
-    ("@SHLIBDIR@", "@LLVM_LIBS_DIR@", config.environment.get(shlibpath_var, ""))
+    (config.shlibdir, config.llvm_libs_dir, config.environment.get(shlibpath_var, ""))
 )
 
 # It is not realistically possible to account for all options that could

clang-tools-extra/test/Unit/lit.cfg.py

@@ -21,7 +21,7 @@
 
 if platform.system() == "Darwin":
     shlibpath_var = "DYLD_LIBRARY_PATH"
-elif platform.system() == "Windows":
+elif platform.system() == "Windows" or sys.platform == "cygwin":
     shlibpath_var = "PATH"
 else:
     shlibpath_var = "LD_LIBRARY_PATH"

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -13676,6 +13676,9 @@ def err_acc_reduction_recipe_no_op
             "not have a valid operation available">;
 def note_acc_reduction_recipe_noop_field
     : Note<"while forming combiner for compound type %0">;
+def note_acc_reduction_combiner_forming
+    : Note<"while forming %select{|binary operator '%1'|conditional "
+           "operator|final assignment operator}0">;
 
 // AMDGCN builtins diagnostics
 def err_amdgcn_load_lds_size_invalid_value : Error<"invalid size value">;

clang/include/clang/CIR/Dialect/Builder/CIRBaseBuilder.h

@@ -311,8 +311,9 @@ class CIRBaseBuilderTy : public mlir::OpBuilder {
   }
 
   /// Create a copy with inferred length.
-  cir::CopyOp createCopy(mlir::Value dst, mlir::Value src) {
-    return cir::CopyOp::create(*this, dst.getLoc(), dst, src);
+  cir::CopyOp createCopy(mlir::Value dst, mlir::Value src,
+                         bool isVolatile = false) {
+    return cir::CopyOp::create(*this, dst.getLoc(), dst, src, isVolatile);
   }
 
   cir::StoreOp createStore(mlir::Location loc, mlir::Value val, mlir::Value dst,

clang/include/clang/CIR/Dialect/IR/CIROps.td

@@ -2724,6 +2724,8 @@ def CIR_CopyOp : CIR_Op<"copy",[
     type of `src` and `dst` must match and both must implement the
     `DataLayoutTypeInterface`.
 
+    The `volatile` keyword indicates that the operation is volatile.
+
     Examples:
 
     ```mlir
@@ -2734,10 +2736,11 @@ def CIR_CopyOp : CIR_Op<"copy",[
 
   let arguments = (ins
       Arg<CIR_PointerType, "", [MemWrite]>:$dst,
-      Arg<CIR_PointerType, "", [MemRead]>:$src
+      Arg<CIR_PointerType, "", [MemRead]>:$src,
+      UnitAttr:$is_volatile
   );
 
-  let assemblyFormat = [{$src `to` $dst
+  let assemblyFormat = [{$src `to` $dst (`volatile` $is_volatile^)?
                         attr-dict `:` qualified(type($dst))
   }];
   let hasVerifier = 1;

clang/include/clang/CIR/MissingFeatures.h

@@ -177,9 +177,10 @@ struct MissingFeatures {
   static bool atomicInfo() { return false; }
   static bool atomicInfoGetAtomicPointer() { return false; }
   static bool atomicInfoGetAtomicAddress() { return false; }
-  static bool atomicUseLibCall() { return false; }
   static bool atomicScope() { return false; }
   static bool atomicSyncScopeID() { return false; }
+  static bool atomicTypes() { return false; }
+  static bool atomicUseLibCall() { return false; }
 
   // Global ctor handling
   static bool globalCtorLexOrder() { return false; }

clang/lib/Analysis/FlowSensitive/Models/UncheckedOptionalAccessModel.cpp

@@ -241,9 +241,9 @@ auto nulloptTypeDecl() {
 auto hasNulloptType() { return hasType(nulloptTypeDecl()); }
 
 auto inPlaceClass() {
-  return recordDecl(hasAnyName("std::in_place_t", "absl::in_place_t",
-                               "base::in_place_t", "folly::in_place_t",
-                               "bsl::in_place_t"));
+  return namedDecl(hasAnyName("std::in_place_t", "absl::in_place_t",
+                              "base::in_place_t", "folly::in_place_t",
+                              "bsl::in_place_t"));
 }
 
 auto isOptionalNulloptConstructor() {

clang/lib/CIR/CodeGen/CIRGenDecl.cpp

@@ -740,6 +740,16 @@ struct CallStackRestore final : EHScopeStack::Cleanup {
 };
 } // namespace
 
+/// Push the standard destructor for the given type as
+/// at least a normal cleanup.
+void CIRGenFunction::pushDestroy(QualType::DestructionKind dtorKind,
+                                 Address addr, QualType type) {
+  assert(dtorKind && "cannot push destructor for trivial type");
+
+  CleanupKind cleanupKind = getCleanupKind(dtorKind);
+  pushDestroy(cleanupKind, addr, type, getDestroyer(dtorKind));
+}
+
 void CIRGenFunction::pushDestroy(CleanupKind cleanupKind, Address addr,
                                  QualType type, Destroyer *destroyer) {
   pushFullExprCleanup<DestroyObject>(cleanupKind, addr, type, destroyer);

clang/lib/CIR/CodeGen/CIRGenExpr.cpp

@@ -1626,14 +1626,15 @@ LValue CIRGenFunction::emitBinaryOperatorLValue(const BinaryOperator *e) {
 
 /// Emit code to compute the specified expression which
 /// can have any type.  The result is returned as an RValue struct.
-RValue CIRGenFunction::emitAnyExpr(const Expr *e, AggValueSlot aggSlot) {
+RValue CIRGenFunction::emitAnyExpr(const Expr *e, AggValueSlot aggSlot,
+                                   bool ignoreResult) {
   switch (CIRGenFunction::getEvaluationKind(e->getType())) {
   case cir::TEK_Scalar:
     return RValue::get(emitScalarExpr(e));
   case cir::TEK_Complex:
     return RValue::getComplex(emitComplexExpr(e));
   case cir::TEK_Aggregate: {
-    if (aggSlot.isIgnored())
+    if (!ignoreResult && aggSlot.isIgnored())
       aggSlot = createAggTemp(e->getType(), getLoc(e->getSourceRange()),
                               getCounterAggTmpAsString());
     emitAggExpr(e, aggSlot);
@@ -1869,8 +1870,7 @@ RValue CIRGenFunction::emitCallExpr(const clang::CallExpr *e,
 /// Emit code to compute the specified expression, ignoring the result.
 void CIRGenFunction::emitIgnoredExpr(const Expr *e) {
   if (e->isPRValue()) {
-    assert(!cir::MissingFeatures::aggValueSlot());
-    emitAnyExpr(e);
+    emitAnyExpr(e, AggValueSlot::ignored(), /*ignoreResult=*/true);
     return;
   }
 

clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp

@@ -24,6 +24,73 @@ using namespace clang;
 using namespace clang::CIRGen;
 
 namespace {
+// FIXME(cir): This should be a common helper between CIRGen
+// and traditional CodeGen
+/// Is the value of the given expression possibly a reference to or
+/// into a __block variable?
+static bool isBlockVarRef(const Expr *e) {
+  // Make sure we look through parens.
+  e = e->IgnoreParens();
+
+  // Check for a direct reference to a __block variable.
+  if (const DeclRefExpr *dre = dyn_cast<DeclRefExpr>(e)) {
+    const VarDecl *var = dyn_cast<VarDecl>(dre->getDecl());
+    return (var && var->hasAttr<BlocksAttr>());
+  }
+
+  // More complicated stuff.
+
+  // Binary operators.
+  if (const BinaryOperator *op = dyn_cast<BinaryOperator>(e)) {
+    // For an assignment or pointer-to-member operation, just care
+    // about the LHS.
+    if (op->isAssignmentOp() || op->isPtrMemOp())
+      return isBlockVarRef(op->getLHS());
+
+    // For a comma, just care about the RHS.
+    if (op->getOpcode() == BO_Comma)
+      return isBlockVarRef(op->getRHS());
+
+    // FIXME: pointer arithmetic?
+    return false;
+
+    // Check both sides of a conditional operator.
+  } else if (const AbstractConditionalOperator *op =
+                 dyn_cast<AbstractConditionalOperator>(e)) {
+    return isBlockVarRef(op->getTrueExpr()) ||
+           isBlockVarRef(op->getFalseExpr());
+
+    // OVEs are required to support BinaryConditionalOperators.
+  } else if (const OpaqueValueExpr *op = dyn_cast<OpaqueValueExpr>(e)) {
+    if (const Expr *src = op->getSourceExpr())
+      return isBlockVarRef(src);
+
+    // Casts are necessary to get things like (*(int*)&var) = foo().
+    // We don't really care about the kind of cast here, except
+    // we don't want to look through l2r casts, because it's okay
+    // to get the *value* in a __block variable.
+  } else if (const CastExpr *cast = dyn_cast<CastExpr>(e)) {
+    if (cast->getCastKind() == CK_LValueToRValue)
+      return false;
+    return isBlockVarRef(cast->getSubExpr());
+
+    // Handle unary operators.  Again, just aggressively look through
+    // it, ignoring the operation.
+  } else if (const UnaryOperator *uop = dyn_cast<UnaryOperator>(e)) {
+    return isBlockVarRef(uop->getSubExpr());
+
+    // Look into the base of a field access.
+  } else if (const MemberExpr *mem = dyn_cast<MemberExpr>(e)) {
+    return isBlockVarRef(mem->getBase());
+
+    // Look into the base of a subscript.
+  } else if (const ArraySubscriptExpr *sub = dyn_cast<ArraySubscriptExpr>(e)) {
+    return isBlockVarRef(sub->getBase());
+  }
+
+  return false;
+}
+
 class AggExprEmitter : public StmtVisitor<AggExprEmitter> {
 
   CIRGenFunction &cgf;
@@ -41,9 +108,7 @@ class AggExprEmitter : public StmtVisitor<AggExprEmitter> {
   AggValueSlot ensureSlot(mlir::Location loc, QualType t) {
     if (!dest.isIgnored())
       return dest;
-
-    cgf.cgm.errorNYI(loc, "Slot for ignored address");
-    return dest;
+    return cgf.createAggTemp(t, loc, "agg.tmp.ensured");
   }
 
   void ensureDest(mlir::Location loc, QualType ty) {
@@ -89,6 +154,47 @@ class AggExprEmitter : public StmtVisitor<AggExprEmitter> {
     (void)cgf.emitCompoundStmt(*e->getSubStmt(), &retAlloca, dest);
   }
 
+  void VisitBinAssign(const BinaryOperator *e) {
+    // For an assignment to work, the value on the right has
+    // to be compatible with the value on the left.
+    assert(cgf.getContext().hasSameUnqualifiedType(e->getLHS()->getType(),
+                                                   e->getRHS()->getType()) &&
+           "Invalid assignment");
+
+    if (isBlockVarRef(e->getLHS()) &&
+        e->getRHS()->HasSideEffects(cgf.getContext())) {
+      cgf.cgm.errorNYI(e->getSourceRange(),
+                       "block var reference with side effects");
+      return;
+    }
+
+    LValue lhs = cgf.emitLValue(e->getLHS());
+
+    // If we have an atomic type, evaluate into the destination and then
+    // do an atomic copy.
+    assert(!cir::MissingFeatures::atomicTypes());
+
+    // Codegen the RHS so that it stores directly into the LHS.
+    assert(!cir::MissingFeatures::aggValueSlotGC());
+    AggValueSlot lhsSlot = AggValueSlot::forLValue(
+        lhs, AggValueSlot::IsDestructed, AggValueSlot::IsAliased,
+        AggValueSlot::MayOverlap);
+
+    // A non-volatile aggregate destination might have volatile member.
+    if (!lhsSlot.isVolatile() && cgf.hasVolatileMember(e->getLHS()->getType()))
+      lhsSlot.setVolatile(true);
+
+    cgf.emitAggExpr(e->getRHS(), lhsSlot);
+
+    // Copy into the destination if the assignment isn't ignored.
+    emitFinalDestCopy(e->getType(), lhs);
+
+    if (!dest.isIgnored() && !dest.isExternallyDestructed() &&
+        e->getType().isDestructedType() == QualType::DK_nontrivial_c_struct)
+      cgf.pushDestroy(QualType::DK_nontrivial_c_struct, dest.getAddress(),
+                      e->getType());
+  }
+
   void VisitDeclRefExpr(DeclRefExpr *e) { emitAggLoadOfLValue(e); }
 
   void VisitInitListExpr(InitListExpr *e);
@@ -186,9 +292,6 @@ class AggExprEmitter : public StmtVisitor<AggExprEmitter> {
     cgf.cgm.errorNYI(e->getSourceRange(),
                      "AggExprEmitter: VisitPointerToDataMemberBinaryOperator");
   }
-  void VisitBinAssign(const BinaryOperator *e) {
-    cgf.cgm.errorNYI(e->getSourceRange(), "AggExprEmitter: VisitBinAssign");
-  }
   void VisitBinComma(const BinaryOperator *e) {
     cgf.emitIgnoredExpr(e->getLHS());
     Visit(e->getRHS());
@@ -503,7 +606,8 @@ void AggExprEmitter::emitCopy(QualType type, const AggValueSlot &dest,
   LValue destLV = cgf.makeAddrLValue(dest.getAddress(), type);
   LValue srcLV = cgf.makeAddrLValue(src.getAddress(), type);
   assert(!cir::MissingFeatures::aggValueSlotVolatile());
-  cgf.emitAggregateCopy(destLV, srcLV, type, dest.mayOverlap());
+  cgf.emitAggregateCopy(destLV, srcLV, type, dest.mayOverlap(),
+                        dest.isVolatile() || src.isVolatile());
 }
 
 void AggExprEmitter::emitInitializationToLValue(Expr *e, LValue lv) {
@@ -804,7 +908,8 @@ void CIRGenFunction::emitAggExpr(const Expr *e, AggValueSlot slot) {
 }
 
 void CIRGenFunction::emitAggregateCopy(LValue dest, LValue src, QualType ty,
-                                       AggValueSlot::Overlap_t mayOverlap) {
+                                       AggValueSlot::Overlap_t mayOverlap,
+                                       bool isVolatile) {
   // TODO(cir): this function needs improvements, commented code for now since
   // this will be touched again soon.
   assert(!ty->isAnyComplexType() && "Unexpected copy of complex");
@@ -860,7 +965,7 @@ void CIRGenFunction::emitAggregateCopy(LValue dest, LValue src, QualType ty,
     cgm.errorNYI("emitAggregateCopy: GC");
 
   [[maybe_unused]] cir::CopyOp copyOp =
-      builder.createCopy(destPtr.getPointer(), srcPtr.getPointer());
+      builder.createCopy(destPtr.getPointer(), srcPtr.getPointer(), isVolatile);
 
   assert(!cir::MissingFeatures::opTBAA());
 }