[Clang][C++23] Implement P1774R8: Portable assumptions (#81014)

This implements the C++23 `[[assume]]` attribute.

Assumption information is lowered to a call to `@llvm.assume`, unless the expression has side-effects, in which case it is discarded and a warning is issued to tell the user that the assumption doesn’t do anything. A failed assumption at compile time is an error (unless we are in `MSVCCompat` mode, in which case we don’t check assumptions at compile time).

Due to performance regressions in LLVM, assumptions can be disabled with the `-fno-assumptions` flag. With it, assumptions will still be parsed and checked, but no calls to `@llvm.assume` will be emitted and assumptions will not be checked at compile time.

Author: Sirraide

clang/docs/ReleaseNotes.rst

@@ -100,6 +100,7 @@ C++23 Feature Support
 
 - Implemented `P2718R0: Lifetime extension in range-based for loops <https://wg21.link/P2718R0>`_. Also
   materialize temporary object which is a prvalue in discarded-value expression.
+- Implemented `P1774R8: Portable assumptions <https://wg21.link/P1774R8>`_.
 
 - Implemented `P2448R2: Relaxing some constexpr restrictions <https://wg21.link/P2448R2>`_.
 

clang/include/clang/Basic/Attr.td

@@ -1580,6 +1580,13 @@ def Unlikely : StmtAttr {
 }
 def : MutualExclusions<[Likely, Unlikely]>;
 
+def CXXAssume : StmtAttr {
+  let Spellings = [CXX11<"", "assume", 202207>];
+  let Subjects = SubjectList<[NullStmt], ErrorDiag, "empty statements">;
+  let Args = [ExprArgument<"Assumption">];
+  let Documentation = [CXXAssumeDocs];
+}
+
 def NoMerge : DeclOrStmtAttr {
   let Spellings = [Clang<"nomerge">];
   let Documentation = [NoMergeDocs];
@@ -4151,11 +4158,11 @@ def OMPDeclareVariant : InheritableAttr {
   }];
 }
 
-def Assumption : InheritableAttr {
+def OMPAssume : InheritableAttr {
   let Spellings = [Clang<"assume">];
   let Subjects = SubjectList<[Function, ObjCMethod]>;
   let InheritEvenIfAlreadyPresent = 1;
-  let Documentation = [AssumptionDocs];
+  let Documentation = [OMPAssumeDocs];
   let Args = [StringArgument<"Assumption">];
 }
 

clang/include/clang/Basic/AttrDocs.td

@@ -1996,6 +1996,34 @@ Here is an example:
   }];
 }
 
+def CXXAssumeDocs : Documentation {
+  let Category = DocCatStmt;
+  let Heading = "assume";
+  let Content = [{
+The ``assume`` attribute is used to indicate to the optimizer that a
+certain condition is assumed to be true at a certain point in the
+program. If this condition is violated at runtime, the behavior is
+undefined. ``assume`` can only be applied to a null statement.
+
+Different optimisers are likely to react differently to the presence of
+this attribute; in some cases, adding ``assume`` may affect performance
+negatively. It should be used with parsimony and care.
+
+Note that `clang::assume` is a different attribute. Always write ``assume``
+without a namespace if you intend to use the standard C++ attribute.
+
+Example:
+
+.. code-block:: c++
+
+  int f(int x, int y) {
+    [[assume(x == 27)]];
+    [[assume(x == y)]];
+    return y + 1; // May be optimised to `return 28`.
+  }
+  }];
+}
+
 def LikelihoodDocs : Documentation {
   let Category = DocCatStmt;
   let Heading = "likely and unlikely";
@@ -4629,7 +4657,7 @@ For more information see
 }];
 }
 
-def AssumptionDocs : Documentation {
+def OMPAssumeDocs : Documentation {
   let Category = DocCatFunction;
   let Heading = "assume";
   let Content = [{

clang/include/clang/Basic/DiagnosticASTKinds.td

@@ -399,6 +399,8 @@ def note_constexpr_unsupported_flexible_array : Note<
   "flexible array initialization is not yet supported">;
 def note_constexpr_non_const_vectorelements : Note<
   "cannot determine number of elements for sizeless vectors in a constant expression">;
+def note_constexpr_assumption_failed : Note<
+  "assumption evaluated to false">;
 def err_experimental_clang_interp_failed : Error<
   "the experimental clang interpreter failed to evaluate an expression">;
 

clang/include/clang/Basic/DiagnosticGroups.td

@@ -1133,9 +1133,11 @@ def NonGCC : DiagGroup<"non-gcc",
 def CXX14Attrs : DiagGroup<"c++14-attribute-extensions">;
 def CXX17Attrs : DiagGroup<"c++17-attribute-extensions">;
 def CXX20Attrs : DiagGroup<"c++20-attribute-extensions">;
+def CXX23Attrs : DiagGroup<"c++23-attribute-extensions">;
 def FutureAttrs : DiagGroup<"future-attribute-extensions", [CXX14Attrs,
                                                             CXX17Attrs,
-                                                            CXX20Attrs]>;
+                                                            CXX20Attrs,
+                                                            CXX23Attrs]>;
 
 def CXX23AttrsOnLambda : DiagGroup<"c++23-lambda-attributes">;
 

clang/include/clang/Basic/DiagnosticParseKinds.td

@@ -786,6 +786,9 @@ def err_ms_property_expected_comma_or_rparen : Error<
 def err_ms_property_initializer : Error<
   "property declaration cannot have a default member initializer">;
 
+def err_assume_attr_expects_cond_expr : Error<
+  "use of this expression in an %0 attribute requires parentheses">;
+
 def warn_cxx20_compat_explicit_bool : Warning<
   "this expression will be parsed as explicit(bool) in C++20">,
   InGroup<CXX20Compat>, DefaultIgnore;

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -855,10 +855,10 @@ def note_strncat_wrong_size : Note<
 def warn_assume_side_effects : Warning<
   "the argument to %0 has side effects that will be discarded">,
   InGroup<DiagGroup<"assume">>;
-def warn_assume_attribute_string_unknown : Warning<
+def warn_omp_assume_attribute_string_unknown : Warning<
   "unknown assumption string '%0'; attribute is potentially ignored">,
   InGroup<UnknownAssumption>;
-def warn_assume_attribute_string_unknown_suggested : Warning<
+def warn_omp_assume_attribute_string_unknown_suggested : Warning<
   "unknown assumption string '%0' may be misspelled; attribute is potentially "
   "ignored, did you mean '%1'?">,
   InGroup<MisspelledAssumption>;
@@ -9115,6 +9115,8 @@ def ext_cxx17_attr : Extension<
   "use of the %0 attribute is a C++17 extension">, InGroup<CXX17Attrs>;
 def ext_cxx20_attr : Extension<
   "use of the %0 attribute is a C++20 extension">, InGroup<CXX20Attrs>;
+def ext_cxx23_attr : Extension<
+  "use of the %0 attribute is a C++23 extension">, InGroup<CXX23Attrs>;
 
 def warn_unused_comparison : Warning<
   "%select{equality|inequality|relational|three-way}0 comparison result unused">,
@@ -10169,6 +10171,9 @@ def err_fallthrough_attr_outside_switch : Error<
 def err_fallthrough_attr_invalid_placement : Error<
   "fallthrough annotation does not directly precede switch label">;
 
+def err_assume_attr_args : Error<
+  "attribute '%0' requires a single expression argument">;
+
 def warn_unreachable_default : Warning<
   "default label in switch which covers all enumeration values">,
   InGroup<CoveredSwitchDefault>, DefaultIgnore;

clang/include/clang/Basic/LangOptions.def

@@ -450,6 +450,8 @@ LANGOPT(RegCall4, 1, 0, "Set __regcall4 as a default calling convention to respe
 
 LANGOPT(MatrixTypes, 1, 0, "Enable or disable the builtin matrix type")
 
+LANGOPT(CXXAssumptions, 1, 1, "Enable or disable codegen and compile-time checks for C++23's [[assume]] attribute")
+
 ENUM_LANGOPT(StrictFlexArraysLevel, StrictFlexArraysLevelKind, 2,
              StrictFlexArraysLevelKind::Default,
              "Rely on strict definition of flexible arrays")

clang/include/clang/Driver/Options.td

@@ -3789,6 +3789,12 @@ def foptimization_record_passes_EQ : Joined<["-"], "foptimization-record-passes=
   HelpText<"Only include passes which match a specified regular expression in the generated optimization record (by default, include all passes)">,
   MetaVarName<"<regex>">;
 
+defm assumptions : BoolFOption<"assumptions",
+  LangOpts<"CXXAssumptions">, DefaultTrue,
+  NegFlag<SetFalse, [], [ClangOption, CC1Option],
+          "Disable codegen and compile-time checks for C++23's [[assume]] attribute">,
+  PosFlag<SetTrue>>;
+
 def fvectorize : Flag<["-"], "fvectorize">, Group<f_Group>,
   HelpText<"Enable the loop vectorization passes">;
 def fno_vectorize : Flag<["-"], "fno-vectorize">, Group<f_Group>;

clang/include/clang/Parse/Parser.h

@@ -1803,6 +1803,7 @@ class Parser : public CodeCompletionHandler {
   ExprResult ParseConstraintLogicalOrExpression(bool IsTrailingRequiresClause);
   // Expr that doesn't include commas.
   ExprResult ParseAssignmentExpression(TypeCastState isTypeCast = NotTypeCast);
+  ExprResult ParseConditionalExpression();
 
   ExprResult ParseMSAsmIdentifier(llvm::SmallVectorImpl<Token> &LineToks,
                                   unsigned &NumLineToksConsumed,
@@ -2955,6 +2956,12 @@ class Parser : public CodeCompletionHandler {
                                SourceLocation ScopeLoc,
                                CachedTokens &OpenMPTokens);
 
+  /// Parse a C++23 assume() attribute. Returns true on error.
+  bool ParseCXXAssumeAttributeArg(ParsedAttributes &Attrs,
+                                  IdentifierInfo *AttrName,
+                                  SourceLocation AttrNameLoc,
+                                  SourceLocation *EndLoc);
+
   IdentifierInfo *TryParseCXX11AttributeIdentifier(
       SourceLocation &Loc,
       Sema::AttributeCompletion Completion = Sema::AttributeCompletion::None,

clang/include/clang/Sema/Sema.h

@@ -9011,6 +9011,12 @@ class Sema final {
   void ProcessStmtAttributes(Stmt *Stmt, const ParsedAttributes &InAttrs,
                              SmallVectorImpl<const Attr *> &OutAttrs);
 
+  ExprResult ActOnCXXAssumeAttr(Stmt *St, const ParsedAttr &A,
+                                SourceRange Range);
+  ExprResult BuildCXXAssumeExpr(Expr *Assumption,
+                                const IdentifierInfo *AttrName,
+                                SourceRange Range);
+
   ///@}
 
   //
@@ -14716,10 +14722,10 @@ class Sema final {
   SmallVector<OMPDeclareVariantScope, 4> OMPDeclareVariantScopes;
 
   /// The current `omp begin/end assumes` scopes.
-  SmallVector<AssumptionAttr *, 4> OMPAssumeScoped;
+  SmallVector<OMPAssumeAttr *, 4> OMPAssumeScoped;
 
   /// All `omp assumes` we encountered so far.
-  SmallVector<AssumptionAttr *, 4> OMPAssumeGlobal;
+  SmallVector<OMPAssumeAttr *, 4> OMPAssumeGlobal;
 
   /// OMPD_loop is mapped to OMPD_for, OMPD_distribute or OMPD_simd depending
   /// on the parameter of the bind clause. In the methods for the

clang/lib/AST/ExprConstant.cpp

@@ -5582,6 +5582,29 @@ static EvalStmtResult EvaluateStmt(StmtResult &Result, EvalInfo &Info,
     MSConstexprContextRAII ConstexprContext(
         *Info.CurrentCall, hasSpecificAttr<MSConstexprAttr>(AS->getAttrs()) &&
                                isa<ReturnStmt>(SS));
+
+    auto LO = Info.getCtx().getLangOpts();
+    if (LO.CXXAssumptions && !LO.MSVCCompat) {
+      for (auto *Attr : AS->getAttrs()) {
+        auto *AA = dyn_cast<CXXAssumeAttr>(Attr);
+        if (!AA)
+          continue;
+
+        auto *Assumption = AA->getAssumption();
+        if (Assumption->isValueDependent())
+          return ESR_Failed;
+
+        bool Value;
+        if (!EvaluateAsBooleanCondition(Assumption, Value, Info))
+          return ESR_Failed;
+        if (!Value) {
+          Info.CCEDiag(Assumption->getExprLoc(),
+                       diag::note_constexpr_assumption_failed);
+          return ESR_Failed;
+        }
+      }
+    }
+
     return EvaluateStmt(Result, Info, SS, Case);
   }
 

clang/lib/CodeGen/CGCall.cpp

@@ -1796,14 +1796,14 @@ static void AddAttributesFromFunctionProtoType(ASTContext &Ctx,
     FuncAttrs.addAttribute("aarch64_inout_zt0");
 }
 
-static void AddAttributesFromAssumes(llvm::AttrBuilder &FuncAttrs,
-                                     const Decl *Callee) {
+static void AddAttributesFromOMPAssumes(llvm::AttrBuilder &FuncAttrs,
+                                        const Decl *Callee) {
   if (!Callee)
     return;
 
   SmallVector<StringRef, 4> Attrs;
 
-  for (const AssumptionAttr *AA : Callee->specific_attrs<AssumptionAttr>())
+  for (const OMPAssumeAttr *AA : Callee->specific_attrs<OMPAssumeAttr>())
     AA->getAssumption().split(Attrs, ",");
 
   if (!Attrs.empty())
@@ -2344,7 +2344,7 @@ void CodeGenModule::ConstructAttributeList(StringRef Name,
 
   // Attach assumption attributes to the declaration. If this is a call
   // site, attach assumptions from the caller to the call as well.
-  AddAttributesFromAssumes(FuncAttrs, TargetDecl);
+  AddAttributesFromOMPAssumes(FuncAttrs, TargetDecl);
 
   bool HasOptnone = false;
   // The NoBuiltinAttr attached to the target FunctionDecl.

clang/lib/CodeGen/CGStmt.cpp

@@ -728,11 +728,19 @@ void CodeGenFunction::EmitAttributedStmt(const AttributedStmt &S) {
     case attr::AlwaysInline:
       alwaysinline = true;
       break;
-    case attr::MustTail:
+    case attr::MustTail: {
       const Stmt *Sub = S.getSubStmt();
       const ReturnStmt *R = cast<ReturnStmt>(Sub);
       musttail = cast<CallExpr>(R->getRetValue()->IgnoreParens());
-      break;
+    } break;
+    case attr::CXXAssume: {
+      const Expr *Assumption = cast<CXXAssumeAttr>(A)->getAssumption();
+      if (getLangOpts().CXXAssumptions &&
+          !Assumption->HasSideEffects(getContext())) {
+        llvm::Value *AssumptionVal = EvaluateExprAsBool(Assumption);
+        Builder.CreateAssumption(AssumptionVal);
+      }
+    } break;
     }
   }
   SaveAndRestore save_nomerge(InNoMergeAttributedStmt, nomerge);

clang/lib/Driver/ToolChains/Clang.cpp

@@ -6982,6 +6982,11 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                         (!IsWindowsMSVC || IsMSVC2015Compatible)))
     CmdArgs.push_back("-fno-threadsafe-statics");
 
+  // Add -fno-assumptions, if it was specified.
+  if (!Args.hasFlag(options::OPT_fassumptions, options::OPT_fno_assumptions,
+                    true))
+    CmdArgs.push_back("-fno-assumptions");
+
   // -fgnu-keywords default varies depending on language; only pass if
   // specified.
   Args.AddLastArg(CmdArgs, options::OPT_fgnu_keywords,

clang/lib/Parse/ParseDeclCXX.cpp

@@ -4528,6 +4528,61 @@ static bool IsBuiltInOrStandardCXX11Attribute(IdentifierInfo *AttrName,
   }
 }
 
+/// Parse the argument to C++23's [[assume()]] attribute.
+bool Parser::ParseCXXAssumeAttributeArg(ParsedAttributes &Attrs,
+                                        IdentifierInfo *AttrName,
+                                        SourceLocation AttrNameLoc,
+                                        SourceLocation *EndLoc) {
+  assert(Tok.is(tok::l_paren) && "Not a C++11 attribute argument list");
+  BalancedDelimiterTracker T(*this, tok::l_paren);
+  T.consumeOpen();
+
+  // [dcl.attr.assume]: The expression is potentially evaluated.
+  EnterExpressionEvaluationContext Unevaluated(
+      Actions, Sema::ExpressionEvaluationContext::PotentiallyEvaluated);
+
+  TentativeParsingAction TPA(*this);
+  ExprResult Res(
+      Actions.CorrectDelayedTyposInExpr(ParseConditionalExpression()));
+  if (Res.isInvalid()) {
+    TPA.Commit();
+    SkipUntil(tok::r_paren, tok::r_square, StopAtSemi | StopBeforeMatch);
+    if (Tok.is(tok::r_paren))
+      T.consumeClose();
+    return true;
+  }
+
+  if (!Tok.isOneOf(tok::r_paren, tok::r_square)) {
+    // Emit a better diagnostic if this is an otherwise valid expression that
+    // is not allowed here.
+    TPA.Revert();
+    Res = ParseExpression();
+    if (!Res.isInvalid()) {
+      auto *E = Res.get();
+      Diag(E->getExprLoc(), diag::err_assume_attr_expects_cond_expr)
+          << AttrName << FixItHint::CreateInsertion(E->getBeginLoc(), "(")
+          << FixItHint::CreateInsertion(PP.getLocForEndOfToken(E->getEndLoc()),
+                                        ")")
+          << E->getSourceRange();
+    }
+
+    T.consumeClose();
+    return true;
+  }
+
+  TPA.Commit();
+  ArgsUnion Assumption = Res.get();
+  auto RParen = Tok.getLocation();
+  T.consumeClose();
+  Attrs.addNew(AttrName, SourceRange(AttrNameLoc, RParen), nullptr,
+               SourceLocation(), &Assumption, 1, ParsedAttr::Form::CXX11());
+
+  if (EndLoc)
+    *EndLoc = RParen;
+
+  return false;
+}
+
 /// ParseCXX11AttributeArgs -- Parse a C++11 attribute-argument-clause.
 ///
 /// [C++11] attribute-argument-clause:
@@ -4596,7 +4651,12 @@ bool Parser::ParseCXX11AttributeArgs(
   if (ScopeName && (ScopeName->isStr("clang") || ScopeName->isStr("_Clang")))
     NumArgs = ParseClangAttributeArgs(AttrName, AttrNameLoc, Attrs, EndLoc,
                                       ScopeName, ScopeLoc, Form);
-  else
+  // So does C++23's assume() attribute.
+  else if (!ScopeName && AttrName->isStr("assume")) {
+    if (ParseCXXAssumeAttributeArg(Attrs, AttrName, AttrNameLoc, EndLoc))
+      return true;
+    NumArgs = 1;
+  } else
     NumArgs = ParseAttributeArgsCommon(AttrName, AttrNameLoc, Attrs, EndLoc,
                                        ScopeName, ScopeLoc, Form);
 

clang/lib/Parse/ParseExpr.cpp

@@ -179,6 +179,19 @@ ExprResult Parser::ParseAssignmentExpression(TypeCastState isTypeCast) {
   return ParseRHSOfBinaryExpression(LHS, prec::Assignment);
 }
 
+ExprResult Parser::ParseConditionalExpression() {
+  if (Tok.is(tok::code_completion)) {
+    cutOffParsing();
+    Actions.CodeCompleteExpression(getCurScope(),
+                                   PreferredType.get(Tok.getLocation()));
+    return ExprError();
+  }
+
+  ExprResult LHS = ParseCastExpression(
+      AnyCastExpr, /*isAddressOfOperand=*/false, NotTypeCast);
+  return ParseRHSOfBinaryExpression(LHS, prec::Conditional);
+}
+
 /// Parse an assignment expression where part of an Objective-C message
 /// send has already been parsed.
 ///