[clang] [Clang] VectorExprEvaluator::VisitCallExpr / InterpretBuiltin - allow AVX/AVX512 subvector extraction intrinsics to be used in constexpr #157712 (PR #162836)

[via cfe-commits]

================
@@ -2858,6 +2858,92 @@ static bool interp__builtin_elementwise_triop(
   return true;
 }
 
+static bool interp__builtin_x86_extract_vector(InterpState &S, CodePtr OpPC,
+                                                 const CallExpr *Call,
+                                                 unsigned ID) {
+  assert(Call->getNumArgs() == 2);
+
+  APSInt ImmAPS = popToAPSInt(S, Call->getArg(1));
+  uint64_t Index = ImmAPS.getZExtValue();
+
+  const Pointer &Src = S.Stk.pop<Pointer>();
+  if (!Src.getFieldDesc()->isPrimitiveArray())
+    return false;
+
+  const Pointer &Dst = S.Stk.peek<Pointer>();
+  if (!Dst.getFieldDesc()->isPrimitiveArray())
+    return false;
+
+  unsigned SrcElems = Src.getNumElems();
+  unsigned DstElems = Dst.getNumElems();
+
+  if (SrcElems == 0 || DstElems == 0 || (SrcElems % DstElems) != 0)
+    return false;
----------------
SeongjaeP wrote:

You’re right, the vector lengths are fixed by definition, so this check isn’t needed. Thanks for catching that!

https://github.com/llvm/llvm-project/pull/162836