merge main into amd-staging

Author: ronlieb

clang/lib/CIR/Dialect/IR/CIRDialect.cpp

@@ -330,6 +330,12 @@ static LogicalResult checkConstantTypes(mlir::Operation *op, mlir::Type opType,
         "zero expects struct, array, vector, or complex type");
   }
 
+  if (mlir::isa<cir::UndefAttr>(attrType)) {
+    if (!mlir::isa<cir::VoidType>(opType))
+      return success();
+    return op->emitOpError("undef expects non-void type");
+  }
+
   if (mlir::isa<cir::BoolAttr>(attrType)) {
     if (!mlir::isa<cir::BoolType>(opType))
       return op->emitOpError("result type (")

clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp

@@ -240,7 +240,7 @@ class CIRAttrToValue {
         .Case<cir::IntAttr, cir::FPAttr, cir::ConstComplexAttr,
               cir::ConstArrayAttr, cir::ConstRecordAttr, cir::ConstVectorAttr,
               cir::ConstPtrAttr, cir::GlobalViewAttr, cir::TypeInfoAttr,
-              cir::VTableAttr, cir::ZeroAttr>(
+              cir::UndefAttr, cir::VTableAttr, cir::ZeroAttr>(
             [&](auto attrT) { return visitCirAttr(attrT); })
         .Default([&](auto attrT) { return mlir::Value(); });
   }
@@ -254,6 +254,7 @@ class CIRAttrToValue {
   mlir::Value visitCirAttr(cir::ConstVectorAttr attr);
   mlir::Value visitCirAttr(cir::GlobalViewAttr attr);
   mlir::Value visitCirAttr(cir::TypeInfoAttr attr);
+  mlir::Value visitCirAttr(cir::UndefAttr attr);
   mlir::Value visitCirAttr(cir::VTableAttr attr);
   mlir::Value visitCirAttr(cir::ZeroAttr attr);
 
@@ -591,6 +592,13 @@ mlir::Value CIRAttrToValue::visitCirAttr(cir::TypeInfoAttr typeInfoAttr) {
   return result;
 }
 
+/// UndefAttr visitor.
+mlir::Value CIRAttrToValue::visitCirAttr(cir::UndefAttr undefAttr) {
+  mlir::Location loc = parentOp->getLoc();
+  return mlir::LLVM::UndefOp::create(
+      rewriter, loc, converter->convertType(undefAttr.getType()));
+}
+
 // VTableAttr visitor.
 mlir::Value CIRAttrToValue::visitCirAttr(cir::VTableAttr vtableArr) {
   mlir::Type llvmTy = converter->convertType(vtableArr.getType());
@@ -2046,9 +2054,11 @@ CIRToLLVMGlobalOpLowering::matchAndRewriteRegionInitializedGlobal(
     cir::GlobalOp op, mlir::Attribute init,
     mlir::ConversionPatternRewriter &rewriter) const {
   // TODO: Generalize this handling when more types are needed here.
-  assert((isa<cir::ConstArrayAttr, cir::ConstRecordAttr, cir::ConstVectorAttr,
-              cir::ConstPtrAttr, cir::ConstComplexAttr, cir::GlobalViewAttr,
-              cir::TypeInfoAttr, cir::VTableAttr, cir::ZeroAttr>(init)));
+  assert(
+      (isa<cir::ConstArrayAttr, cir::ConstRecordAttr, cir::ConstVectorAttr,
+           cir::ConstPtrAttr, cir::ConstComplexAttr, cir::GlobalViewAttr,
+           cir::TypeInfoAttr, cir::UndefAttr, cir::VTableAttr, cir::ZeroAttr>(
+          init)));
 
   // TODO(cir): once LLVM's dialect has proper equivalent attributes this
   // should be updated. For now, we use a custom op to initialize globals
@@ -2106,8 +2116,8 @@ mlir::LogicalResult CIRToLLVMGlobalOpLowering::matchAndRewrite(
     } else if (mlir::isa<cir::ConstArrayAttr, cir::ConstVectorAttr,
                          cir::ConstRecordAttr, cir::ConstPtrAttr,
                          cir::ConstComplexAttr, cir::GlobalViewAttr,
-                         cir::TypeInfoAttr, cir::VTableAttr, cir::ZeroAttr>(
-                   init.value())) {
+                         cir::TypeInfoAttr, cir::UndefAttr, cir::VTableAttr,
+                         cir::ZeroAttr>(init.value())) {
       // TODO(cir): once LLVM's dialect has proper equivalent attributes this
       // should be updated. For now, we use a custom op to initialize globals
       // to the appropriate value.

clang/lib/CodeGen/BackendUtil.cpp

@@ -1134,6 +1134,8 @@ void EmitAssemblyHelper::RunOptimizationPipeline(
                   CodeGenOpts.SanitizeMinimalRuntime),
               /*MayReturn=*/
               CodeGenOpts.SanitizeRecover.has(SanitizerKind::LocalBounds),
+              /*HandlerPreserveAllRegs=*/
+              static_cast<bool>(CodeGenOpts.SanitizeHandlerPreserveAllRegs),
           };
         }
         FPM.addPass(BoundsCheckingPass(Options));

clang/lib/CodeGen/CGExpr.cpp

@@ -3819,6 +3819,8 @@ static void emitCheckHandlerCall(CodeGenFunction &CGF,
   bool NeedsAbortSuffix =
       IsFatal && RecoverKind != CheckRecoverableKind::Unrecoverable;
   bool MinimalRuntime = CGF.CGM.getCodeGenOpts().SanitizeMinimalRuntime;
+  bool HandlerPreserveAllRegs =
+      CGF.CGM.getCodeGenOpts().SanitizeHandlerPreserveAllRegs;
   const SanitizerHandlerInfo &CheckInfo = SanitizerHandlers[CheckHandler];
   const StringRef CheckName = CheckInfo.Name;
   std::string FnName = "__ubsan_handle_" + CheckName.str();
@@ -3828,6 +3830,8 @@ static void emitCheckHandlerCall(CodeGenFunction &CGF,
     FnName += "_minimal";
   if (NeedsAbortSuffix)
     FnName += "_abort";
+  if (HandlerPreserveAllRegs && !NeedsAbortSuffix)
+    FnName += "_preserve";
   bool MayReturn =
       !IsFatal || RecoverKind == CheckRecoverableKind::AlwaysRecoverable;
 
@@ -3848,6 +3852,10 @@ static void emitCheckHandlerCall(CodeGenFunction &CGF,
             (CGF.CurCodeDecl && CGF.CurCodeDecl->hasAttr<OptimizeNoneAttr>());
   if (NoMerge)
     HandlerCall->addFnAttr(llvm::Attribute::NoMerge);
+  if (HandlerPreserveAllRegs && !NeedsAbortSuffix) {
+    // N.B. there is also a clang::CallingConv which is not what we want here.
+    HandlerCall->setCallingConv(llvm::CallingConv::PreserveAll);
+  }
   if (!MayReturn) {
     HandlerCall->setDoesNotReturn();
     CGF.Builder.CreateUnreachable();

clang/lib/CodeGen/CGPointerAuth.cpp

@@ -440,9 +440,10 @@ CodeGenModule::getConstantSignedPointer(llvm::Constant *Pointer, unsigned Key,
     IntegerDiscriminator = llvm::ConstantInt::get(Int64Ty, 0);
   }
 
-  return llvm::ConstantPtrAuth::get(Pointer,
-                                    llvm::ConstantInt::get(Int32Ty, Key),
-                                    IntegerDiscriminator, AddressDiscriminator);
+  return llvm::ConstantPtrAuth::get(
+      Pointer, llvm::ConstantInt::get(Int32Ty, Key), IntegerDiscriminator,
+      AddressDiscriminator,
+      /*DeactivationSymbol=*/llvm::Constant::getNullValue(DefaultPtrTy));
 }
 
 /// Does a given PointerAuthScheme require us to sign a value

clang/lib/Driver/SanitizerArgs.cpp

@@ -419,14 +419,16 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
   const Driver &D = TC.getDriver();
   SanitizerMask TrappingKinds = parseSanitizeTrapArgs(D, Args, DiagnoseErrors);
   SanitizerMask InvalidTrappingKinds = TrappingKinds & NotAllowedWithTrap;
+  const llvm::Triple &Triple = TC.getTriple();
 
   MinimalRuntime =
       Args.hasFlag(options::OPT_fsanitize_minimal_runtime,
                    options::OPT_fno_sanitize_minimal_runtime, MinimalRuntime);
   HandlerPreserveAllRegs =
       Args.hasFlag(options::OPT_fsanitize_handler_preserve_all_regs,
                    options::OPT_fno_sanitize_handler_preserve_all_regs,
-                   HandlerPreserveAllRegs);
+                   HandlerPreserveAllRegs) &&
+      MinimalRuntime && (Triple.isAArch64() || Triple.isX86_64());
 
   // The object size sanitizer should not be enabled at -O0.
   Arg *OptLevel = Args.getLastArg(options::OPT_O_Group);
@@ -494,7 +496,6 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
       // -fsanitize=function and -fsanitize=kcfi instrument indirect function
       // calls to load a type hash before the function label. Therefore, an
       // execute-only target doesn't support the function and kcfi sanitizers.
-      const llvm::Triple &Triple = TC.getTriple();
       if (isExecuteOnlyTarget(Triple, Args)) {
         if (SanitizerMask KindsToDiagnose =
                 Add & NotAllowedWithExecuteOnly & ~DiagnosedKinds) {

clang/lib/Driver/ToolChains/Linux.cpp

@@ -1156,7 +1156,7 @@ SanitizerMask Linux::getSupportedSanitizers() const {
   if (IsX86_64 || IsMIPS64 || IsAArch64 || IsPowerPC64 || IsSystemZ ||
       IsLoongArch64 || IsRISCV64)
     Res |= SanitizerKind::Thread;
-  if (IsX86_64 || IsAArch64 || IsSystemZ)
+  if (IsX86_64 || IsAArch64)
     Res |= SanitizerKind::Type;
   if (IsX86_64 || IsSystemZ || IsPowerPC64)
     Res |= SanitizerKind::KernelMemory;

clang/test/CIR/CodeGen/lambda.cpp

@@ -8,6 +8,39 @@
 // We declare anonymous record types to represent lambdas. Rather than trying to
 // to match the declarations, we establish variables for these when they are used.
 
+auto global_lambda = [](){};
+void use_global_lambda() {
+  global_lambda();
+}
+
+// CIR: cir.global "private" internal dso_local @global_lambda = #cir.undef : ![[REC_LAM_GLOBAL_LAMBDA:.*]] {alignment = 1 : i64}
+// CIR: cir.func lambda internal private dso_local @_ZNK3$_0clEv(%[[THIS_ARG:.*]]: !cir.ptr<![[REC_LAM_GLOBAL_LAMBDA]]> {{.*}})
+// CIR:   %[[THIS:.*]] = cir.alloca !cir.ptr<![[REC_LAM_GLOBAL_LAMBDA]]>, !cir.ptr<!cir.ptr<![[REC_LAM_GLOBAL_LAMBDA]]>>, ["this", init]
+// CIR:   cir.store %[[THIS_ARG]], %[[THIS]]
+// CIR:   cir.load %[[THIS]]
+//
+// CIR: cir.func {{.*}} @_Z17use_global_lambdav()
+// CIR:   %[[LAMBDA:.*]] = cir.get_global @global_lambda : !cir.ptr<![[REC_LAM_GLOBAL_LAMBDA]]>
+// CIR:   cir.call @_ZNK3$_0clEv(%[[LAMBDA]]) : (!cir.ptr<![[REC_LAM_GLOBAL_LAMBDA]]>) -> ()
+
+// LLVM: @global_lambda = internal global %[[REC_LAM_GLOBAL_LAMBDA:.*]] undef, align 1
+// LLVM: define internal void @"_ZNK3$_0clEv"(ptr %[[THIS_ARG:.*]])
+// LLVM:   %[[THIS_ADDR:.*]] = alloca ptr
+// LLVM:   store ptr %[[THIS_ARG]], ptr %[[THIS_ADDR]]
+// LLVM:   %[[THIS:.*]] = load ptr, ptr %[[THIS_ADDR]]
+//
+// LLVM: define dso_local void @_Z17use_global_lambdav()
+// LLVM:   call void @"_ZNK3$_0clEv"(ptr @global_lambda)
+
+// OGCG: @global_lambda = internal global %[[REC_LAM_GLOBAL_LAMBDA:.*]] undef, align 1
+// OGCG: define dso_local void @_Z17use_global_lambdav()
+// OGCG:   call void @"_ZNK3$_0clEv"(ptr noundef nonnull align 1 dereferenceable(1) @global_lambda)
+//
+// OGCG: define internal void @"_ZNK3$_0clEv"(ptr {{.*}} %[[THIS_ARG:.*]])
+// OGCG:   %[[THIS_ADDR:.*]] = alloca ptr
+// OGCG:   store ptr %[[THIS_ARG]], ptr %[[THIS_ADDR]]
+// OGCG:   %[[THIS:.*]] = load ptr, ptr %[[THIS_ADDR]]
+
 void fn() {
   auto a = [](){};
   a();