Merged main:511ba45a47d6 into amd-gfx:8a11303f2b4e

Local branch amd-gfx 8a11303 Merged main:586986a063ee into amd-gfx:b8966b19ced0
Remote branch main 511ba45 [X86][MC][CodeGen] Support EGPR for KMOV (llvm#73781)

Author: SC llvm team

clang/docs/ReleaseNotes.rst

@@ -793,6 +793,11 @@ Bug Fixes to C++ Support
 - Fix crash when parsing nested requirement. Fixes:
   (`#73112 <https://github.com/llvm/llvm-project/issues/73112>`_)
 
+- Clang now immediately instantiates function template specializations
+  at the end of the definition of the corresponding function template
+  when the definition appears after the first point of instantiation.
+  (`#73232 <https://github.com/llvm/llvm-project/issues/73232>`_)
+
 Bug Fixes to AST Handling
 ^^^^^^^^^^^^^^^^^^^^^^^^^
 - Fixed an import failure of recursive friend class template.

clang/include/clang/Sema/ExternalSemaSource.h

@@ -181,6 +181,9 @@ class ExternalSemaSource : public ExternalASTSource {
                  SmallVectorImpl<std::pair<ValueDecl *,
                                            SourceLocation> > &Pending) {}
 
+  virtual void ReadPendingInstantiationsOfConstexprEntity(
+      const NamedDecl *D, llvm::SmallSetVector<NamedDecl *, 4> &Decls){};
+
   /// Read the set of late parsed template functions for this source.
   ///
   /// The external source should insert its own late parsed template functions

clang/include/clang/Sema/MultiplexExternalSemaSource.h

@@ -319,6 +319,9 @@ class MultiplexExternalSemaSource : public ExternalSemaSource {
   void ReadPendingInstantiations(
      SmallVectorImpl<std::pair<ValueDecl*, SourceLocation> >& Pending) override;
 
+  virtual void ReadPendingInstantiationsOfConstexprEntity(
+      const NamedDecl *D, llvm::SmallSetVector<NamedDecl *, 4> &Decls) override;
+
   /// Read the set of late parsed template functions for this source.
   ///
   /// The external source should insert its own late parsed template functions

clang/include/clang/Sema/Sema.h

@@ -59,6 +59,7 @@
 #include "clang/Sema/TypoCorrection.h"
 #include "clang/Sema/Weak.h"
 #include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallBitVector.h"
 #include "llvm/ADT/SmallPtrSet.h"
@@ -10087,6 +10088,12 @@ class Sema final {
   /// but have not yet been performed.
   std::deque<PendingImplicitInstantiation> PendingInstantiations;
 
+  /// Track constexpr functions referenced before they are (lexically) defined.
+  /// The key is the pattern, associated with a list of specialisations that
+  /// need to be instantiated when the pattern is defined.
+  llvm::DenseMap<NamedDecl *, SmallVector<NamedDecl *>>
+      PendingInstantiationsOfConstexprEntities;
+
   /// Queue of implicit template instantiations that cannot be performed
   /// eagerly.
   SmallVector<PendingImplicitInstantiation, 1> LateParsedInstantiations;
@@ -10405,6 +10412,9 @@ class Sema final {
                                      bool Recursive = false,
                                      bool DefinitionRequired = false,
                                      bool AtEndOfTU = false);
+
+  void PerformPendingInstantiationsOfConstexprFunctions(FunctionDecl *Template);
+
   VarTemplateSpecializationDecl *BuildVarTemplateInstantiation(
       VarTemplateDecl *VarTemplate, VarDecl *FromVar,
       const TemplateArgumentList &TemplateArgList,

clang/include/clang/Serialization/ASTBitCodes.h

@@ -695,6 +695,10 @@ enum ASTRecordTypes {
   /// Record code for an unterminated \#pragma clang assume_nonnull begin
   /// recorded in a preamble.
   PP_ASSUME_NONNULL_LOC = 67,
+
+  /// Record code for constexpr templated entities that have been used but not
+  /// yet instantiated.
+  PENDING_INSTANTIATIONS_OF_CONSTEXPR_ENTITIES = 68,
 };
 
 /// Record types used within a source manager block.

clang/include/clang/Serialization/ASTReader.h

@@ -814,6 +814,9 @@ class ASTReader
   /// is the instantiation location.
   SmallVector<serialization::DeclID, 64> PendingInstantiations;
 
+  llvm::DenseMap<serialization::DeclID, std::set<serialization::DeclID>>
+      PendingInstantiationsOfConstexprEntities;
+
   //@}
 
   /// \name DiagnosticsEngine-relevant special data
@@ -2101,6 +2104,9 @@ class ASTReader
                   SmallVectorImpl<std::pair<ValueDecl *,
                                             SourceLocation>> &Pending) override;
 
+  virtual void ReadPendingInstantiationsOfConstexprEntity(
+      const NamedDecl *D, llvm::SmallSetVector<NamedDecl *, 4> &Decls) override;
+
   void ReadLateParsedTemplates(
       llvm::MapVector<const FunctionDecl *, std::unique_ptr<LateParsedTemplate>>
           &LPTMap) override;

clang/lib/Sema/MultiplexExternalSemaSource.cpp

@@ -310,6 +310,12 @@ void MultiplexExternalSemaSource::ReadPendingInstantiations(
     Sources[i]->ReadPendingInstantiations(Pending);
 }
 
+void MultiplexExternalSemaSource::ReadPendingInstantiationsOfConstexprEntity(
+    const NamedDecl *D, llvm::SmallSetVector<NamedDecl *, 4> &Decls) {
+  for (size_t i = 0; i < Sources.size(); ++i)
+    Sources[i]->ReadPendingInstantiationsOfConstexprEntity(D, Decls);
+}
+
 void MultiplexExternalSemaSource::ReadLateParsedTemplates(
     llvm::MapVector<const FunctionDecl *, std::unique_ptr<LateParsedTemplate>>
         &LPTMap) {

clang/lib/Sema/SemaDecl.cpp

@@ -16275,6 +16275,9 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
   if (FD && !FD->isDeleted())
     checkTypeSupport(FD->getType(), FD->getLocation(), FD);
 
+  if (FD && FD->isConstexpr() && FD->isTemplated())
+    PerformPendingInstantiationsOfConstexprFunctions(FD);
+
   return dcl;
 }
 

clang/lib/Sema/SemaExpr.cpp

@@ -19053,12 +19053,17 @@ void Sema::MarkFunctionReferenced(SourceLocation Loc, FunctionDecl *Func,
               CodeSynthesisContexts.size())
             PendingLocalImplicitInstantiations.push_back(
                 std::make_pair(Func, PointOfInstantiation));
-          else if (Func->isConstexpr())
+          else if (Func->isConstexpr()) {
             // Do not defer instantiations of constexpr functions, to avoid the
             // expression evaluator needing to call back into Sema if it sees a
             // call to such a function.
             InstantiateFunctionDefinition(PointOfInstantiation, Func);
-          else {
+            if (!Func->isDefined()) {
+              PendingInstantiationsOfConstexprEntities
+                  [Func->getTemplateInstantiationPattern()->getCanonicalDecl()]
+                      .push_back(Func);
+            }
+          } else {
             Func->setInstantiationIsPending(true);
             PendingInstantiations.push_back(
                 std::make_pair(Func, PointOfInstantiation));

clang/lib/Sema/SemaTemplateInstantiateDecl.cpp

@@ -6495,6 +6495,34 @@ void Sema::PerformPendingInstantiations(bool LocalOnly) {
     PendingInstantiations.swap(delayedPCHInstantiations);
 }
 
+// Instantiate all referenced specializations of the given function template
+// definition. This make sure that constexpr function templates that are defined
+// after the point of instantiation of their use can be evaluated after they
+// are defined. see CWG2497.
+void Sema::PerformPendingInstantiationsOfConstexprFunctions(FunctionDecl *Tpl) {
+
+  auto InstantiateAll = [&](const auto &Range) {
+    for (NamedDecl *D : Range) {
+      FunctionDecl *Fun = cast<FunctionDecl>(D);
+      InstantiateFunctionDefinition(Fun->getPointOfInstantiation(), Fun);
+    }
+  };
+
+  auto It =
+      PendingInstantiationsOfConstexprEntities.find(Tpl->getCanonicalDecl());
+  if (It != PendingInstantiationsOfConstexprEntities.end()) {
+    auto Decls = std::move(It->second);
+    PendingInstantiationsOfConstexprEntities.erase(It);
+    InstantiateAll(Decls);
+  }
+
+  llvm::SmallSetVector<NamedDecl *, 4> Decls;
+  if (ExternalSource) {
+    ExternalSource->ReadPendingInstantiationsOfConstexprEntity(Tpl, Decls);
+    InstantiateAll(Decls);
+  }
+}
+
 void Sema::PerformDependentDiagnostics(const DeclContext *Pattern,
                        const MultiLevelTemplateArgumentList &TemplateArgs) {
   for (auto *DD : Pattern->ddiags()) {

clang/lib/Serialization/ASTReader.cpp

@@ -3709,6 +3709,19 @@ llvm::Error ASTReader::ReadASTBlock(ModuleFile &F,
       }
       break;
 
+    case PENDING_INSTANTIATIONS_OF_CONSTEXPR_ENTITIES:
+      if (Record.size() % 2 != 0)
+        return llvm::createStringError(
+            std::errc::illegal_byte_sequence,
+            "Invalid PENDING_INSTANTIATIONS_OF_CONSTEXPR_ENTITIES block");
+
+      for (unsigned I = 0, N = Record.size(); I != N; /* in loop */) {
+        DeclID Key = getGlobalDeclID(F, Record[I++]);
+        DeclID Value = getGlobalDeclID(F, Record[I++]);
+        PendingInstantiationsOfConstexprEntities[Key].insert(Value);
+      }
+      break;
+
     case SEMA_DECL_REFS:
       if (Record.size() != 3)
         return llvm::createStringError(std::errc::illegal_byte_sequence,
@@ -8718,6 +8731,20 @@ void ASTReader::ReadPendingInstantiations(
   PendingInstantiations.clear();
 }
 
+void ASTReader::ReadPendingInstantiationsOfConstexprEntity(
+    const NamedDecl *D, llvm::SmallSetVector<NamedDecl *, 4> &Decls) {
+  for (auto *Redecl : D->redecls()) {
+    if (!Redecl->isFromASTFile())
+      continue;
+    DeclID Id = Redecl->getGlobalID();
+    auto It = PendingInstantiationsOfConstexprEntities.find(Id);
+    if (It == PendingInstantiationsOfConstexprEntities.end())
+      continue;
+    for (DeclID InstantiationId : It->second)
+      Decls.insert(cast<NamedDecl>(GetDecl(InstantiationId)));
+  }
+}
+
 void ASTReader::ReadLateParsedTemplates(
     llvm::MapVector<const FunctionDecl *, std::unique_ptr<LateParsedTemplate>>
         &LPTMap) {

clang/lib/Serialization/ASTWriter.cpp

@@ -849,6 +849,7 @@ void ASTWriter::WriteBlockInfoBlock() {
   RECORD(SEMA_DECL_REFS);
   RECORD(WEAK_UNDECLARED_IDENTIFIERS);
   RECORD(PENDING_IMPLICIT_INSTANTIATIONS);
+  RECORD(PENDING_INSTANTIATIONS_OF_CONSTEXPR_ENTITIES);
   RECORD(UPDATE_VISIBLE);
   RECORD(DECL_UPDATE_OFFSETS);
   RECORD(DECL_UPDATES);
@@ -4836,6 +4837,16 @@ ASTFileSignature ASTWriter::WriteASTCore(Sema &SemaRef, StringRef isysroot,
   assert(SemaRef.PendingLocalImplicitInstantiations.empty() &&
          "There are local ones at end of translation unit!");
 
+  // Build a record containing all pending instantiations of constexpr
+  // entities.
+  RecordData PendingInstantiationsOfConstexprEntities;
+  for (const auto &I : SemaRef.PendingInstantiationsOfConstexprEntities) {
+    for (const auto &Elem : I.second) {
+      AddDeclRef(I.first, PendingInstantiationsOfConstexprEntities);
+      AddDeclRef(Elem, PendingInstantiationsOfConstexprEntities);
+    }
+  }
+
   // Build a record containing some declaration references.
   RecordData SemaDeclRefs;
   if (SemaRef.StdNamespace || SemaRef.StdBadAlloc || SemaRef.StdAlignValT) {
@@ -5153,6 +5164,11 @@ ASTFileSignature ASTWriter::WriteASTCore(Sema &SemaRef, StringRef isysroot,
   if (!PendingInstantiations.empty())
     Stream.EmitRecord(PENDING_IMPLICIT_INSTANTIATIONS, PendingInstantiations);
 
+  // Write the record containing pending instantiations of constexpr entities.
+  if (!PendingInstantiationsOfConstexprEntities.empty())
+    Stream.EmitRecord(PENDING_INSTANTIATIONS_OF_CONSTEXPR_ENTITIES,
+                      PendingInstantiationsOfConstexprEntities);
+
   // Write the record containing declaration references of Sema.
   if (!SemaDeclRefs.empty())
     Stream.EmitRecord(SEMA_DECL_REFS, SemaDeclRefs);

clang/test/PCH/instantiate-used-constexpr-function.cpp

@@ -0,0 +1,17 @@
+// RUN: %clang_cc1 -std=c++2a -emit-pch %s -o %t
+// RUN: %clang_cc1 -std=c++2a -include-pch %t -verify %s
+
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+template<typename T> constexpr T f();
+constexpr int g() { return f<int>(); } // #1
+
+#else /*included pch*/
+
+template<typename T> constexpr T f() { return 123; }
+int k[g()];
+
+#endif // HEADER

clang/test/SemaTemplate/instantiate-used-constexpr-function.cpp

@@ -0,0 +1,30 @@
+// RUN: %clang_cc1 -fsyntax-only -verify %s
+// expected-no-diagnostics
+
+namespace GH73232  {
+
+template <typename _CharT>
+struct basic_string {
+  constexpr void _M_construct();
+  constexpr basic_string() {
+    _M_construct();
+  }
+};
+
+basic_string<char> a;
+
+template <typename _CharT>
+constexpr void basic_string<_CharT>::_M_construct(){}
+constexpr basic_string<char> str{};
+
+template <typename T>
+constexpr void g(T);
+
+constexpr int f() { g(0); return 0; }
+
+template <typename T>
+constexpr void g(T) {}
+
+constexpr int z = f();
+
+}

llvm/include/llvm/Config/llvm-config.h.cmake

@@ -16,7 +16,7 @@
 
 /* Indicate that this is LLVM compiled from the amd-gfx branch. */
 #define LLVM_HAVE_BRANCH_AMD_GFX
-#define LLVM_MAIN_REVISION 482223
+#define LLVM_MAIN_REVISION 482228
 
 /* Define if LLVM_ENABLE_DUMP is enabled */
 #cmakedefine LLVM_ENABLE_DUMP

llvm/include/llvm/IR/IntrinsicsAMDGPU.td

@@ -2024,12 +2024,14 @@ def int_amdgcn_alignbyte : ClangBuiltin<"__builtin_amdgcn_alignbyte">,
   [IntrNoMem, IntrSpeculatable]
 >;
 
-def int_amdgcn_mul_i24 : DefaultAttrsIntrinsic<[llvm_i32_ty],
+// mul24 intrinsics can return i32 or i64.
+// When returning i64, they're lowered to a mul24/mulhi24 pair.
+def int_amdgcn_mul_i24 : DefaultAttrsIntrinsic<[llvm_anyint_ty],
   [llvm_i32_ty, llvm_i32_ty],
   [IntrNoMem, IntrSpeculatable]
 >;
 
-def int_amdgcn_mul_u24 : DefaultAttrsIntrinsic<[llvm_i32_ty],
+def int_amdgcn_mul_u24 : DefaultAttrsIntrinsic<[llvm_anyint_ty],
   [llvm_i32_ty, llvm_i32_ty],
   [IntrNoMem, IntrSpeculatable]
 >;

llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp

@@ -624,34 +624,6 @@ static Value *insertValues(IRBuilder<> &Builder,
   return NewVal;
 }
 
-// Returns 24-bit or 48-bit (as per `NumBits` and `Size`) mul of `LHS` and
-// `RHS`. `NumBits` is the number of KnownBits of the result and `Size` is the
-// width of the original destination.
-static Value *getMul24(IRBuilder<> &Builder, Value *LHS, Value *RHS,
-                       unsigned Size, unsigned NumBits, bool IsSigned) {
-  if (Size <= 32 || NumBits <= 32) {
-    Intrinsic::ID ID =
-        IsSigned ? Intrinsic::amdgcn_mul_i24 : Intrinsic::amdgcn_mul_u24;
-    return Builder.CreateIntrinsic(ID, {}, {LHS, RHS});
-  }
-
-  assert(NumBits <= 48);
-
-  Intrinsic::ID LoID =
-      IsSigned ? Intrinsic::amdgcn_mul_i24 : Intrinsic::amdgcn_mul_u24;
-  Intrinsic::ID HiID =
-      IsSigned ? Intrinsic::amdgcn_mulhi_i24 : Intrinsic::amdgcn_mulhi_u24;
-
-  Value *Lo = Builder.CreateIntrinsic(LoID, {}, {LHS, RHS});
-  Value *Hi = Builder.CreateIntrinsic(HiID, {}, {LHS, RHS});
-
-  IntegerType *I64Ty = Builder.getInt64Ty();
-  Lo = Builder.CreateZExtOrTrunc(Lo, I64Ty);
-  Hi = Builder.CreateZExtOrTrunc(Hi, I64Ty);
-
-  return Builder.CreateOr(Lo, Builder.CreateShl(Hi, 32));
-}
-
 bool AMDGPUCodeGenPrepareImpl::replaceMulWithMul24(BinaryOperator &I) const {
   if (I.getOpcode() != Instruction::Mul)
     return false;
@@ -691,26 +663,20 @@ bool AMDGPUCodeGenPrepareImpl::replaceMulWithMul24(BinaryOperator &I) const {
   extractValues(Builder, RHSVals, RHS);
 
   IntegerType *I32Ty = Builder.getInt32Ty();
-  for (int I = 0, E = LHSVals.size(); I != E; ++I) {
-    Value *LHS, *RHS;
-    if (IsSigned) {
-      LHS = Builder.CreateSExtOrTrunc(LHSVals[I], I32Ty);
-      RHS = Builder.CreateSExtOrTrunc(RHSVals[I], I32Ty);
-    } else {
-      LHS = Builder.CreateZExtOrTrunc(LHSVals[I], I32Ty);
-      RHS = Builder.CreateZExtOrTrunc(RHSVals[I], I32Ty);
-    }
+  IntegerType *IntrinTy = Size > 32 ? Builder.getInt64Ty() : I32Ty;
+  Type *DstTy = LHSVals[0]->getType();
 
-    Value *Result =
-        getMul24(Builder, LHS, RHS, Size, LHSBits + RHSBits, IsSigned);
-
-    if (IsSigned) {
-      ResultVals.push_back(
-          Builder.CreateSExtOrTrunc(Result, LHSVals[I]->getType()));
-    } else {
-      ResultVals.push_back(
-          Builder.CreateZExtOrTrunc(Result, LHSVals[I]->getType()));
-    }
+  for (int I = 0, E = LHSVals.size(); I != E; ++I) {
+    Value *LHS = IsSigned ? Builder.CreateSExtOrTrunc(LHSVals[I], I32Ty)
+                          : Builder.CreateZExtOrTrunc(LHSVals[I], I32Ty);
+    Value *RHS = IsSigned ? Builder.CreateSExtOrTrunc(RHSVals[I], I32Ty)
+                          : Builder.CreateZExtOrTrunc(RHSVals[I], I32Ty);
+    Intrinsic::ID ID =
+        IsSigned ? Intrinsic::amdgcn_mul_i24 : Intrinsic::amdgcn_mul_u24;
+    Value *Result = Builder.CreateIntrinsic(ID, {IntrinTy}, {LHS, RHS});
+    Result = IsSigned ? Builder.CreateSExtOrTrunc(Result, DstTy)
+                      : Builder.CreateZExtOrTrunc(Result, DstTy);
+    ResultVals.push_back(Result);
   }
 
   Value *NewVal = insertValues(Builder, Ty, ResultVals);