merge main into amd-staging

lands and reverts internally
  d36f66b [NFC][offload][OMPT] Cleanup of OMPT internals (llvm#109005)
downstream gerrit review contains the content we want landed.

Change-Id: I9aebe6a0aa1775d85f7225642bcb6bdc6c5de41f

Author: ronlieb

bolt/include/bolt/Rewrite/RewriteInstance.h

@@ -510,12 +510,11 @@ class RewriteInstance {
   };
 
   /// Different types of X86-64 PLT sections.
-  const PLTSectionInfo X86_64_PLTSections[4] = {
-      { ".plt", 16 },
-      { ".plt.got", 8 },
-      { ".plt.sec", 8 },
-      { nullptr, 0 }
-  };
+  const PLTSectionInfo X86_64_PLTSections[5] = {{".plt", 16},
+                                                {".plt.got", 8},
+                                                {".plt.sec", 8},
+                                                {".iplt", 16},
+                                                {nullptr, 0}};
 
   /// AArch64 PLT sections.
   const PLTSectionInfo AArch64_PLTSections[4] = {

bolt/lib/Rewrite/RewriteInstance.cpp

@@ -1533,7 +1533,7 @@ void RewriteInstance::createPLTBinaryFunction(uint64_t TargetAddress,
 
   MCSymbol *Symbol = Rel->Symbol;
   if (!Symbol) {
-    if (!BC->isAArch64() || !Rel->Addend || !Rel->isIRelative())
+    if (BC->isRISCV() || !Rel->Addend || !Rel->isIRelative())
       return;
 
     // IFUNC trampoline without symbol

bolt/test/AArch64/ifunc.c renamed to bolt/test/AArch64/ifunc.test

@@ -1,8 +1,6 @@
-// This test checks that IFUNC trampoline is properly recognised by BOLT
-
 // With -O0 indirect call is performed on IPLT trampoline. IPLT trampoline
 // has IFUNC symbol.
-// RUN: %clang %cflags -nostdlib -O0 -no-pie %s -fuse-ld=lld \
+// RUN: %clang %cflags -nostdlib -O0 -no-pie %p/../Inputs/ifunc.c -fuse-ld=lld \
 // RUN:    -o %t.O0.exe -Wl,-q
 // RUN: llvm-bolt %t.O0.exe -o %t.O0.bolt.exe \
 // RUN:   --print-disasm --print-only=_start | \
@@ -12,7 +10,7 @@
 
 // Non-pie static executable doesn't generate PT_DYNAMIC, check relocation
 // is readed successfully and IPLT trampoline has been identified by bolt.
-// RUN: %clang %cflags -nostdlib -O3 %s -fuse-ld=lld -no-pie \
+// RUN: %clang %cflags -nostdlib -O3 %p/../Inputs/ifunc.c -fuse-ld=lld -no-pie \
 // RUN:   -o %t.O3_nopie.exe -Wl,-q
 // RUN: llvm-readelf -l %t.O3_nopie.exe | \
 // RUN:   FileCheck --check-prefix=NON_DYN_CHECK %s
@@ -25,7 +23,7 @@
 // With -O3 direct call is performed on IPLT trampoline. IPLT trampoline
 // doesn't have associated symbol. The ifunc symbol has the same address as
 // IFUNC resolver function.
-// RUN: %clang %cflags -nostdlib -O3 %s -fuse-ld=lld -fPIC -pie \
+// RUN: %clang %cflags -nostdlib -O3 %p/../Inputs/ifunc.c -fuse-ld=lld -fPIC -pie \
 // RUN:   -o %t.O3_pie.exe -Wl,-q
 // RUN: llvm-bolt %t.O3_pie.exe -o %t.O3_pie.bolt.exe  \
 // RUN:   --print-disasm --print-only=_start | \
@@ -35,8 +33,8 @@
 
 // Check that IPLT trampoline located in .plt section are normally handled by
 // BOLT. The gnu-ld linker doesn't use separate .iplt section.
-// RUN: %clang %cflags -nostdlib -O3 %s -fuse-ld=lld -fPIC -pie \
-// RUN:   -T %p/Inputs/iplt.ld -o %t.iplt_O3_pie.exe -Wl,-q
+// RUN: %clang %cflags -nostdlib -O3 %p/../Inputs/ifunc.c -fuse-ld=lld -fPIC -pie \
+// RUN:   -T %p/../Inputs/iplt.ld -o %t.iplt_O3_pie.exe -Wl,-q
 // RUN: llvm-bolt %t.iplt_O3_pie.exe -o %t.iplt_O3_pie.bolt.exe  \
 // RUN:   --print-disasm --print-only=_start  | \
 // RUN:   FileCheck --check-prefix=CHECK %s
@@ -49,14 +47,3 @@
 
 // REL_CHECK: R_AARCH64_IRELATIVE [[#%x,REL_SYMB_ADDR:]]
 // REL_CHECK: [[#REL_SYMB_ADDR]] {{.*}} FUNC {{.*}} resolver_foo
-
-static void foo() {}
-static void bar() {}
-
-extern int use_foo;
-
-static void *resolver_foo(void) { return use_foo ? foo : bar; }
-
-__attribute__((ifunc("resolver_foo"))) void ifoo();
-
-void _start() { ifoo(); }

bolt/test/Inputs/ifunc.c

@@ -0,0 +1,12 @@
+// This test checks that IFUNC trampoline is properly recognised by BOLT
+
+static void foo() {}
+static void bar() {}
+
+extern int use_foo;
+
+static void *resolver_foo(void) { return use_foo ? foo : bar; }
+
+__attribute__((ifunc("resolver_foo"))) void ifoo();
+
+void _start() { ifoo(); }

bolt/test/AArch64/Inputs/iplt.ld renamed to bolt/test/Inputs/iplt.ld

@@ -0,0 +1,47 @@
+// Check if BOLT can process ifunc symbols from .plt section
+// RUN: %clang %cflags -nostdlib -no-pie %p/../Inputs/ifunc.c -fuse-ld=lld \
+// RUN:    -o %t.exe -Wl,-q
+// RUN: llvm-bolt %t.exe -o %t.bolt.exe \
+// RUN:   --print-disasm --print-only=_start | \
+// RUN:   FileCheck --check-prefix=CHECK %s
+// RUN: llvm-readelf -aW %t.bolt.exe | \
+// RUN:   FileCheck --check-prefix=REL_CHECK %s
+
+// Check if BOLT can process ifunc symbols from .plt section in non-pie static
+// executable case.
+// RUN: %clang %cflags -nostdlib %p/../Inputs/ifunc.c -fuse-ld=lld -no-pie \
+// RUN:   -o %t.nopie.exe -Wl,-q
+// RUN: llvm-readelf -l %t.nopie.exe | \
+// RUN:   FileCheck --check-prefix=NON_DYN_CHECK %s
+// RUN: llvm-bolt %t.nopie.exe -o %t.nopie.bolt.exe  \
+// RUN:   --print-disasm --print-only=_start | \
+// RUN:   FileCheck --check-prefix=CHECK %s
+// RUN: llvm-readelf -aW %t.nopie.bolt.exe | \
+// RUN:   FileCheck --check-prefix=REL_CHECK %s
+
+// Check if BOLT can process ifunc symbols from .plt section in pie executable
+// case.
+// RUN: %clang %cflags -nostdlib %p/../Inputs/ifunc.c -fuse-ld=lld -fPIC -pie \
+// RUN:   -o %t.pie.exe -Wl,-q
+// RUN: llvm-bolt %t.pie.exe -o %t.pie.bolt.exe  \
+// RUN:   --print-disasm --print-only=_start | \
+// RUN:   FileCheck --check-prefix=CHECK %s
+// RUN: llvm-readelf -aW %t.pie.bolt.exe | \
+// RUN:   FileCheck --check-prefix=REL_CHECK %s
+
+// Check that IPLT trampoline located in .plt section are normally handled by
+// BOLT. The gnu-ld linker doesn't use separate .iplt section.
+// RUN: %clang %cflags -nostdlib %p/../Inputs/ifunc.c -fuse-ld=lld -fPIC -pie \
+// RUN:   -T %p/../Inputs/iplt.ld -o %t.iplt_pie.exe -Wl,-q
+// RUN: llvm-bolt %t.iplt_pie.exe -o %t.iplt_pie.bolt.exe  \
+// RUN:   --print-disasm --print-only=_start  | \
+// RUN:   FileCheck --check-prefix=CHECK %s
+// RUN: llvm-readelf -aW %t.iplt_pie.bolt.exe | \
+// RUN:   FileCheck --check-prefix=REL_CHECK %s
+
+// NON_DYN_CHECK-NOT: DYNAMIC
+
+// CHECK: callq "resolver_foo/1@PLT"
+
+// REL_CHECK: R_X86_64_IRELATIVE [[#%x,REL_SYMB_ADDR:]]
+// REL_CHECK: [[#REL_SYMB_ADDR]] {{.*}} FUNC {{.*}} resolver_foo

bolt/test/X86/ifunc.test

@@ -293,6 +293,11 @@ Attribute Changes in Clang
 - Introduced a new attribute ``[[clang::coro_await_elidable_argument]]`` on function parameters
   to propagate safe elide context to arguments if such function is also under a safe elide context.
 
+- The documentation of the ``[[clang::musttail]]`` attribute was updated to
+  note that the lifetimes of all local variables end before the call. This does
+  not change the behaviour of the compiler, as this was true for previous
+  versions.
+
 Improvements to Clang's diagnostics
 -----------------------------------
 
@@ -349,6 +354,10 @@ Improvements to Clang's diagnostics
 
 - Don't emit bogus dangling diagnostics when ``[[gsl::Owner]]`` and `[[clang::lifetimebound]]` are used together (#GH108272).
 
+- The ``-Wreturn-stack-address`` warning now also warns about addresses of
+  local variables passed to function calls using the ``[[clang::musttail]]``
+  attribute.
+
 Improvements to Clang's time-trace
 ----------------------------------
 

clang/docs/ReleaseNotes.rst

@@ -637,6 +637,12 @@ return value must be trivially destructible. The calling convention of the
 caller and callee must match, and they must not be variadic functions or have
 old style K&R C function declarations.
 
+The lifetimes of all local variables and function parameters end immediately
+before the call to the function. This means that it is undefined behaviour to
+pass a pointer or reference to a local variable to the called function, which
+is not the case without the attribute. Clang will emit a warning in common
+cases where this happens.
+
 ``clang::musttail`` provides assurances that the tail call can be optimized on
 all targets, not just one.
   }];

clang/include/clang/Basic/AttrDocs.td

@@ -10104,11 +10104,15 @@ def err_lifetimebound_ctor_dtor : Error<
 // CHECK: returning address/reference of stack memory
 def warn_ret_stack_addr_ref : Warning<
   "%select{address of|reference to}0 stack memory associated with "
-  "%select{local variable|parameter|compound literal}2 %1 returned">,
+  "%select{local variable|parameter|compound literal}2 %1 "
+  "%select{returned|passed to musttail function}3">,
   InGroup<ReturnStackAddress>;
 def warn_ret_local_temp_addr_ref : Warning<
   "returning %select{address of|reference to}0 local temporary object">,
   InGroup<ReturnStackAddress>;
+def warn_musttail_local_temp_addr_ref : Warning<
+  "passing %select{address of|reference to}0 local temporary object to musttail function">,
+  InGroup<ReturnStackAddress>;
 def err_ret_local_temp_ref : Error<
   "returning reference to local temporary object">;
 def warn_ret_addr_label : Warning<

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -6471,8 +6471,10 @@ def mcx16 : Flag<["-"], "mcx16">, Group<m_x86_Features_Group>;
 def mno_cx16 : Flag<["-"], "mno-cx16">, Group<m_x86_Features_Group>;
 def menqcmd : Flag<["-"], "menqcmd">, Group<m_x86_Features_Group>;
 def mno_enqcmd : Flag<["-"], "mno-enqcmd">, Group<m_x86_Features_Group>;
-def mevex512 : Flag<["-"], "mevex512">, Group<m_x86_Features_Group>;
-def mno_evex512 : Flag<["-"], "mno-evex512">, Group<m_x86_Features_Group>;
+def mevex512 : Flag<["-"], "mevex512">, Group<m_x86_Features_Group>,
+               Visibility<[ClangOption, CLOption, FlangOption]>;
+def mno_evex512 : Flag<["-"], "mno-evex512">, Group<m_x86_Features_Group>,
+                  Visibility<[ClangOption, CLOption, FlangOption]>;
 def mf16c : Flag<["-"], "mf16c">, Group<m_x86_Features_Group>;
 def mno_f16c : Flag<["-"], "mno-f16c">, Group<m_x86_Features_Group>;
 def mfma : Flag<["-"], "mfma">, Group<m_x86_Features_Group>;

clang/include/clang/Driver/Options.td

@@ -53,6 +53,7 @@ class MultiplexConsumer : public SemaConsumer {
 public:
   // Takes ownership of the pointers in C.
   MultiplexConsumer(std::vector<std::unique_ptr<ASTConsumer>> C);
+  MultiplexConsumer(std::unique_ptr<ASTConsumer> C);
   ~MultiplexConsumer() override;
 
   // ASTConsumer
@@ -80,7 +81,7 @@ class MultiplexConsumer : public SemaConsumer {
   void InitializeSema(Sema &S) override;
   void ForgetSema() override;
 
-private:
+protected:
   std::vector<std::unique_ptr<ASTConsumer>> Consumers; // Owns these.
   std::unique_ptr<MultiplexASTMutationListener> MutationListener;
   std::unique_ptr<MultiplexASTDeserializationListener> DeserializationListener;

clang/include/clang/Frontend/MultiplexConsumer.h

@@ -14,11 +14,9 @@
 #ifndef LLVM_CLANG_INTERPRETER_INTERPRETER_H
 #define LLVM_CLANG_INTERPRETER_INTERPRETER_H
 
-#include "clang/AST/Decl.h"
 #include "clang/AST/GlobalDecl.h"
 #include "clang/Interpreter/PartialTranslationUnit.h"
 #include "clang/Interpreter/Value.h"
-#include "clang/Sema/Ownership.h"
 
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ExecutionEngine/JITSymbol.h"
@@ -38,6 +36,9 @@ class ThreadSafeContext;
 namespace clang {
 
 class CompilerInstance;
+class CodeGenerator;
+class CXXRecordDecl;
+class Decl;
 class IncrementalExecutor;
 class IncrementalParser;
 
@@ -77,42 +78,45 @@ class IncrementalCompilerBuilder {
   llvm::StringRef CudaSDKPath;
 };
 
-/// Generate glue code between the Interpreter's built-in runtime and user code.
-class RuntimeInterfaceBuilder {
-public:
-  virtual ~RuntimeInterfaceBuilder() = default;
-
-  using TransformExprFunction = ExprResult(RuntimeInterfaceBuilder *Builder,
-                                           Expr *, ArrayRef<Expr *>);
-  virtual TransformExprFunction *getPrintValueTransformer() = 0;
-};
+class IncrementalAction;
+class InProcessPrintingASTConsumer;
 
 /// Provides top-level interfaces for incremental compilation and execution.
 class Interpreter {
+  friend class Value;
+  friend InProcessPrintingASTConsumer;
+
   std::unique_ptr<llvm::orc::ThreadSafeContext> TSCtx;
+  /// Long-lived, incremental parsing action.
+  std::unique_ptr<IncrementalAction> Act;
   std::unique_ptr<IncrementalParser> IncrParser;
   std::unique_ptr<IncrementalExecutor> IncrExecutor;
-  std::unique_ptr<RuntimeInterfaceBuilder> RuntimeIB;
 
   // An optional parser for CUDA offloading
   std::unique_ptr<IncrementalParser> DeviceParser;
 
+  /// List containing information about each incrementally parsed piece of code.
+  std::list<PartialTranslationUnit> PTUs;
+
   unsigned InitPTUSize = 0;
 
   // This member holds the last result of the value printing. It's a class
   // member because we might want to access it after more inputs. If no value
   // printing happens, it's in an invalid state.
   Value LastValue;
 
-  // Add a call to an Expr to report its result. We query the function from
-  // RuntimeInterfaceBuilder once and store it as a function pointer to avoid
-  // frequent virtual function calls.
-  RuntimeInterfaceBuilder::TransformExprFunction *AddPrintValueCall = nullptr;
+  /// When CodeGen is created the first llvm::Module gets cached in many places
+  /// and we must keep it alive.
+  std::unique_ptr<llvm::Module> CachedInCodeGenModule;
+
+  /// Compiler instance performing the incremental compilation.
+  std::unique_ptr<CompilerInstance> CI;
 
 protected:
   // Derived classes can use an extended interface of the Interpreter.
-  Interpreter(std::unique_ptr<CompilerInstance> CI, llvm::Error &Err,
-              std::unique_ptr<llvm::orc::LLJITBuilder> JITBuilder = nullptr);
+  Interpreter(std::unique_ptr<CompilerInstance> Instance, llvm::Error &Err,
+              std::unique_ptr<llvm::orc::LLJITBuilder> JITBuilder = nullptr,
+              std::unique_ptr<clang::ASTConsumer> Consumer = nullptr);
 
   // Create the internal IncrementalExecutor, or re-create it after calling
   // ResetExecutor().
@@ -122,15 +126,8 @@ class Interpreter {
   // JIT engine. In particular, it doesn't run cleanup or destructors.
   void ResetExecutor();
 
-  // Lazily construct the RuntimeInterfaceBuilder. The provided instance will be
-  // used for the entire lifetime of the interpreter. The default implementation
-  // targets the in-process __clang_Interpreter runtime. Override this to use a
-  // custom runtime.
-  virtual std::unique_ptr<RuntimeInterfaceBuilder> FindRuntimeInterface();
-
 public:
   virtual ~Interpreter();
-
   static llvm::Expected<std::unique_ptr<Interpreter>>
   create(std::unique_ptr<CompilerInstance> CI);
   static llvm::Expected<std::unique_ptr<Interpreter>>
@@ -145,7 +142,6 @@ class Interpreter {
   llvm::Expected<PartialTranslationUnit &> Parse(llvm::StringRef Code);
   llvm::Error Execute(PartialTranslationUnit &T);
   llvm::Error ParseAndExecute(llvm::StringRef Code, Value *V = nullptr);
-  llvm::Expected<llvm::orc::ExecutorAddr> CompileDtorCall(CXXRecordDecl *CXXRD);
 
   /// Undo N previous incremental inputs.
   llvm::Error Undo(unsigned N = 1);
@@ -167,8 +163,6 @@ class Interpreter {
   llvm::Expected<llvm::orc::ExecutorAddr>
   getSymbolAddressFromLinkerName(llvm::StringRef LinkerName) const;
 
-  enum InterfaceKind { NoAlloc, WithAlloc, CopyArray, NewTag };
-
   const llvm::SmallVectorImpl<Expr *> &getValuePrintingInfo() const {
     return ValuePrintingInfo;
   }
@@ -178,7 +172,15 @@ class Interpreter {
 private:
   size_t getEffectivePTUSize() const;
   void markUserCodeStart();
+  llvm::Expected<Expr *> ExtractValueFromExpr(Expr *E);
+  llvm::Expected<llvm::orc::ExecutorAddr> CompileDtorCall(CXXRecordDecl *CXXRD);
+
+  CodeGenerator *getCodeGen() const;
+  std::unique_ptr<llvm::Module> GenModule();
+  PartialTranslationUnit &RegisterPTU(TranslationUnitDecl *TU);
 
+  // A cache for the compiled destructors used to for de-allocation of managed
+  // clang::Values.
   llvm::DenseMap<CXXRecordDecl *, llvm::orc::ExecutorAddr> Dtors;
 
   llvm::SmallVector<Expr *, 4> ValuePrintingInfo;

clang/include/clang/Interpreter/Interpreter.h

@@ -1306,7 +1306,16 @@ static bool interp__builtin_operator_new(InterpState &S, CodePtr OpPC,
     return false;
   }
 
-  // FIXME: CheckArraySize for NumElems?
+  // NB: The same check we're using in CheckArraySize()
+  if (NumElems.getActiveBits() >
+          ConstantArrayType::getMaxSizeBits(S.getASTContext()) ||
+      NumElems.ugt(Descriptor::MaxArrayElemBytes / ElemSize.getQuantity())) {
+    // FIXME: NoThrow check?
+    const SourceInfo &Loc = S.Current->getSource(OpPC);
+    S.FFDiag(Loc, diag::note_constexpr_new_too_large)
+        << NumElems.getZExtValue();
+    return false;
+  }
 
   std::optional<PrimType> ElemT = S.getContext().classify(ElemType);
   DynamicAllocator &Allocator = S.getAllocator();