[clang] [Headers][X86] VectorExprEvaluator::VisitCallExpr - allow SSE/AVX2/AVX512 pack intrinsics to be used in constexpr (PR #156003)

[via cfe-commits]

================
@@ -2604,6 +2605,51 @@ static bool interp__builtin_elementwise_int_binop(
   return true;
 }
 
+static bool interp__builtin_x86_pack(
+    InterpState &S, CodePtr, const CallExpr *E,
+    llvm::function_ref<APSInt(const APSInt &)> narrowElement) {
+  const auto *VT0 = E->getArg(0)->getType()->castAs<VectorType>();
+  const auto *VT1 = E->getArg(1)->getType()->castAs<VectorType>();
+  assert(VT0 && VT1 && "pack builtin VT0 and VT1 must be VectorType");
+  assert(VT0->getElementType() == VT1->getElementType() &&
+         VT0->getNumElements() == VT1->getNumElements() &&
+         "pack builtin VT0 and VT1 ElementType must be same");
+
+  const Pointer &RHS = S.Stk.pop<Pointer>();
+  const Pointer &LHS = S.Stk.pop<Pointer>();
+  const Pointer &Dst = S.Stk.peek<Pointer>();
+
+  ASTContext &Ctx = S.getASTContext();
+  const unsigned SrcBits = Ctx.getIntWidth(VT0->getElementType());
+  const unsigned LHSVecLen = VT0->getNumElements();
+  const unsigned VectorBits = LHSVecLen * SrcBits;
+  const unsigned SrcPerLane = VectorBits >= 128 ? (128 / SrcBits) : LHSVecLen;
+  const unsigned Lanes = VectorBits >= 128 ? (VectorBits / 128) : 1;
+
+  PrimType SrcT = *S.getContext().classify(VT0->getElementType());
+  PrimType DstT = *S.getContext().classify(getElemType(Dst));
----------------
woruyu wrote:

if the two element means VT0 and VT1, they are the same.

https://github.com/llvm/llvm-project/pull/156003