[clang] [X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow element extraction/insertion intrinsics to be used in constexpr #159753 (PR #161302)

[Aditya Chaudhari via cfe-commits]

================
@@ -2878,6 +2878,60 @@ static bool interp__builtin_x86_insert_subvector(InterpState &S, CodePtr OpPC,
   return true;
 }
 
+static bool interp__builtin_vec_ext(InterpState &S, CodePtr OpPC,
+                                    const CallExpr *Call, unsigned ID) {
+  assert(Call->getNumArgs() == 2);
+
+  APSInt ImmAPS = popToAPSInt(S, Call->getArg(1));
+  const Pointer &Vec = S.Stk.pop<Pointer>();
+  if (!Vec.getFieldDesc()->isPrimitiveArray())
+    return false;
+
+  unsigned NumElts = Vec.getNumElems();
+  unsigned Index = static_cast<unsigned>(ImmAPS.getZExtValue() & (NumElts - 1));
+
+  switch (ID) {
----------------
AdityaC4 wrote:

I tried using TYPE_SWITCH, but even with `if constexpr` gurad it expands to pointer/member-pointer cases which don’t have toAPSInt(), so the shared body fails to compile and I get `E No member named 'toAPSInt' in 'clang::interp::MemberPointer' clang (no_member) [2903, 37]` and `No member named 'toAPSInt' in 'clang::interp::Pointer' clang (no_member) [2903, 37]`
I could split the logic: handle PT_Float explicitly and use INT_TYPE_SWITCH_NO_BOOL for the integer cases.


https://github.com/llvm/llvm-project/pull/161302