[clang] [X86][Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow AVX512 mask predicate intrinsics to be used in constexpr (PR #165054)

[Simon Pilgrim via cfe-commits]

================
@@ -3361,6 +3361,57 @@ static bool interp__builtin_ia32_vpconflict(InterpState &S, CodePtr OpPC,
   return true;
 }
 
+static bool interp__builtin_ia32_cvt_mask(InterpState &S, CodePtr OpPC,
+                                          const CallExpr *Call, unsigned ID) {
+  assert(Call->getNumArgs() == 1);
+
+  const Pointer &Vec = S.Stk.pop<Pointer>();
+  APInt RetMask(Vec.getNumElems(), 0);
+  unsigned VectorLen = Vec.getNumElems();
+  PrimType ElemT = Vec.getFieldDesc()->getPrimType();
+
+  for (unsigned ElemNum = 0; ElemNum < VectorLen; ++ElemNum) {
+    APSInt A;
+    INT_TYPE_SWITCH_NO_BOOL(ElemT, { A = Vec.elem<T>(ElemNum).toAPSInt(); });
+    unsigned MSB = A[A.getBitWidth() - 1];
+    RetMask.setBitVal(ElemNum, MSB);
+  }
+  pushInteger(S, RetMask, Call->getType());
+  return true;
+}
+
+static bool interp__builtin_x86_byteshift(
----------------
RKSimon wrote:

get rid of this - we recently replaced interp__builtin_x86_byteshift with the generic shuffle handler and this is now redundant

https://github.com/llvm/llvm-project/pull/165054