[clang][Interp] Handle std::move etc. builtins

[tbaederr]

Only light testing here since I'm not sure how to properly test this.

[llvmbot]

@llvm/pr-subscribers-clang

Author: Timm Baeder (tbaederr)

Changes

Only light testing here since I'm not sure how to properly test this.

---

Full diff: https://github.com/llvm/llvm-project/pull/70772.diff

2 Files Affected:

• (modified) clang/lib/AST/Interp/InterpBuiltin.cpp (+17)
• (modified) clang/test/AST/Interp/functions.cpp (+16)

diff --git a/clang/lib/AST/Interp/InterpBuiltin.cpp b/clang/lib/AST/Interp/InterpBuiltin.cpp
index e329794cb79243d..10e703d868f168a 100644
--- a/clang/lib/AST/Interp/InterpBuiltin.cpp
+++ b/clang/lib/AST/Interp/InterpBuiltin.cpp
@@ -412,6 +412,15 @@ static bool interp__builtin_popcount(InterpState &S, CodePtr OpPC,
   return true;
 }
 
+static bool interp__builtin_move(InterpState &S, CodePtr OpPC,
+                                 const InterpFrame *Frame, const Function *Func,
+                                 const CallExpr *Call) {
+
+  const Pointer &Arg = S.Stk.peek<Pointer>();
+  S.Stk.push<Pointer>(Arg);
+  return Func->getDecl()->isConstexpr();
+}
+
 bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const Function *F,
                       const CallExpr *Call) {
   InterpFrame *Frame = S.Current;
@@ -537,6 +546,14 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const Function *F,
       return retInt(S, OpPC, Dummy);
     break;
 
+  case Builtin::BIas_const:
+  case Builtin::BIforward:
+  case Builtin::BIforward_like:
+  case Builtin::BImove:
+    if (interp__builtin_move(S, OpPC, Frame, F, Call))
+      return Ret<PT_Ptr>(S, OpPC, Dummy);
+    break;
+
   default:
     return false;
   }
diff --git a/clang/test/AST/Interp/functions.cpp b/clang/test/AST/Interp/functions.cpp
index 4bef9c2f7c0d1fa..364744203424380 100644
--- a/clang/test/AST/Interp/functions.cpp
+++ b/clang/test/AST/Interp/functions.cpp
@@ -371,3 +371,19 @@ namespace Variadic {
   constexpr int (*VFP)(...) = variadic_function2;
   static_assert(VFP() == 12, "");
 }
+
+
+namespace std {
+template <typename T> struct remove_reference { using type = T; };
+template <typename T> struct remove_reference<T &> { using type = T; };
+template <typename T> struct remove_reference<T &&> { using type = T; };
+template <typename T>
+constexpr typename std::remove_reference<T>::type&& move(T &&t) noexcept {
+  return static_cast<typename std::remove_reference<T>::type &&>(t);
+}
+}
+/// The std::move declaration above gets translated to a builtin function.
+namespace Move {
+  constexpr int A = std::move(5);
+  static_assert(A == 5, "");
+}

[hnrklssn]

Would making a class with custom move and copy constructors with side effects, and moving/copying an instance around a bit to affect the results of static_asserts, work for testing?

[hnrklssn]

Nice. I realise you're following the convention already established in the test file, but have you considered compiling with -verify=new,both and -verify=ref,both, respectively, and combining the shared diagnostics into both-error etc. to make it easier to see when they do and don't align?

[tbaederr]

Nice. I realise you're following the convention already established in the test file, but have you considered compiling with -verify=new,both and -verify=ref,both, respectively, and combining the shared diagnostics into both-error etc. to make it easier to see when they do and don't align?

That might be an interesting approach for the next new test file I add, thanks.

[tbaederr]

Ping

[tbaederr]

Ping

[hnrklssn]

LGTM, but someone else should approve also.

[tbaederr]

Ping

[tbaederr]

Ping

[AaronBallman]

I'd like to see test coverage involving a move constructor, a move assignment operator, a direct call to __builtin_move, and some testing for std::as_const and std::forward. Of special interest would be times when there's UB in the move constructor/move assignment that should be caught or a non-copyable object where move semantics are the only thing that should work.

[tbaederr]

I've enabled test/SemaCXX/builtin-std-move.cpp with the new interpreter.

[AaronBallman]

LGTM assuming no surprises come up from the additional test coverage.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.