struct AssertTypeEq;

template <typename T>
struct AssertTypeEq<T, T> {
  typedef bool type;
};

// A unique type used as the default value for the arguments of class
// template Types.  This allows us to simulate variadic templates
// (e.g. Types<int>, Type<int, double>, and etc), which C++ doesn't
// support directly.
struct None {};

// The following family of struct and struct templates are used to

struct AssertTypeEq;

template <typename T>
struct AssertTypeEq<T, T> {
  typedef bool type;
};

// A unique type used as the default value for the arguments of class
// template Types.  This allows us to simulate variadic templates
// (e.g. Types<int>, Type<int, double>, and etc), which C++ doesn't
// support directly.
struct None {};

// The following family of struct and struct templates are used to

func.func private @testIfElse(f32) -> f32

// Test invalid tf.If operation
func.func @testInvalidIfOp(tensor<i1>, f32) -> f32 {
^bb0(%arg0: tensor<i1>, %arg1: f32):
  // expected-error @+1 {{operand #1 must be variadic of tensor of tf.dtype values}}
  %1 = "tf.If"(%arg0, %arg1) {
    then_branch = @testIfThen,
    else_branch = @testIfElse,
    is_stateless = false
  } : (tensor<i1>, f32) -> f32

}

// Returns the number of elements in `range`.
template <typename Range>
size_t Size(Range&& range) {
  return range.size();
}

// Returns the total number of elements in a variadic number of `ranges`.
template <typename Range, typename... RangeTs>
size_t Size(Range&& range, RangeTs&&... ranges) {
  return range.size() + Size(std::forward<RangeTs>(ranges)...);
}

//                   |
//               value   [5, 8, 9]
//
//   This can actually be simplified into just:
//
//           =>   Value [5, 8, 9]
// TODO(b/133341698): Move to tablegen when variadic is supported.
struct RemoveRedundantUnpackPack : public RewritePattern {
  explicit RemoveRedundantUnpackPack(MLIRContext* context)
      : RewritePattern(PackOp::getOperationName(), 2, context) {}

      };

  if (inner_op->hasTrait<mlir::OpTrait::AttrSizedOperandSegments>() ||
      inner_op->hasTrait<mlir::OpTrait::AttrSizedResultSegments>()) {
    // The op has multiple variadic operands or results.
    // Calculate operand and result segment sizes using the OpDef.
    NameRangeMap input_ranges, output_ranges;
    // This will fail only if the OpDef is syntactically invalid.

^bb0(%arg0: tensor<? x f16>, %arg1: tensor<? x f16>, %arg2: tensor<? x f16>):
  // expected-error @+1 {{'tfl.add_n' op operand #0 must be variadic of tensor of 32-bit float or 32-bit signless integer}}
  %0 = "tfl.add_n"(%arg0, %arg1, %arg2): (tensor<? x f16>, tensor<? x f16>, tensor<? x f16>) -> tensor<? x f16>
  func.return %0 : tensor<? x f16>
}

// -----

        element_types.push_back(getElementTypeOrSelf(value.getType()));
      }
    }

    if constexpr (std::is_same<DstOpT, mhlo::CaseOp>::value) {
      // Explicitly handle the Case op because it has variadic regions and takes
      // the number of regions as an input along with the operands.
      mhlo_op = rewriter.create<DstOpT>(loc, op.getResultTypes(),
                                        adaptor.getBranchIndex(),

        (incl. anti-aliasing).
    *   Renamed `tf.image` functions to remove duplicate "image" where it is
        redundant.
    *   Variadic reduce is supported on CPU Variadic reduce is supported on CPU
    *   Remove unused `StringViewVariantWrapper`.
    *   Delete unused `Fingerprint64Map` op registration
    *   Add broadcasting support to `tf.matmul`.

	}
	return s.newValue3(ssa.OpSliceMake, n.Type(), p, l, c)
}

// minMax converts an OMIN/OMAX builtin call into SSA.
func (s *state) minMax(n *ir.CallExpr) *ssa.Value {
	// The OMIN/OMAX builtin is variadic, but its semantics are
	// equivalent to left-folding a binary min/max operation across the
	// arguments list.
	fold := func(op func(x, a *ssa.Value) *ssa.Value) *ssa.Value {
		x := s.expr(n.Args[0])