%outputs, %control = tf_executor.island wraps "tf.Const"() {device = "", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>
      %outputs_0, %control_1 = tf_executor.island wraps "tf.Const"() {device = "", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>

  %dense_default_0 = "tf.Const"() {device = "/device:CPU:0", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>
  %dense_default_1 = "tf.Const"() {device = "/device:CPU:0", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>
  %dense_keys = "tf.Const"() {device = "/device:CPU:0", dtype = !tf_type.string, value = dense<""> : tensor<2x!tf_type.string>} : () -> tensor<2x!tf_type.string>

	if tid == 0 {
		return nil, plainError("missing tid")
	}

	f := efaceOf(&fn)
	if f._type == nil || f._type.Kind_&abi.KindMask != abi.Func {
		return nil, plainError("fn must be a function")
	}
	fv := (*funcval)(f.data)

	a := efaceOf(&stackArgs)
	if a._type != nil && a._type.Kind_&abi.KindMask != abi.Pointer {
		return nil, plainError("args must be a pointer or nil")
	}
	argp := a.data

	}
}

// Keep this code in sync with cmd/compile/internal/walk/builtin.go:walkUnsafeSlice
func unsafeslice64(et *_type, ptr unsafe.Pointer, len64 int64) {
	len := int(len64)
	if int64(len) != len64 {
		panicunsafeslicelen1(getcallerpc())
	}
	unsafeslice(et, ptr, len)
}

func unsafeslicecheckptr(et *_type, ptr unsafe.Pointer, len64 int64) {
	unsafeslice64(et, ptr, len64)

  tensorflow::PartialTensorShape unknown_shape;
  TensorSpec ts1;
  ts1.shape = unknown_shape;
  ts1.dtype = tensorflow::DT_FLOAT;

  TensorSpec ts2;
  ts2.shape = tensorflow::PartialTensorShape({2});
  ts2.dtype = tensorflow::DT_FLOAT;

  TensorSpec ts3;
  ts3.shape = tensorflow::PartialTensorShape({1, 2});
  ts3.dtype = tensorflow::DT_FLOAT;

  EXPECT_TRUE(absl::VerifyTypeImplementsAbslHashCorrectly({ts1, ts2, ts3}));
}

      tensorflow::PartialTensorShape* shape) const = 0;

  // Returns tensor (full) type.
  // While there is no immediate plan to deprecate dtype and shape in favor
  // of only using full type type information, this is a future possibility.
  //
  // Note that map_dtype_to_child_of_tensor() from core/framework/types.h
  // can be used to set a FullTypeDef based on dtype in a derived class if
  // appropriate.

  static Status CreateUninitialized(
      ImmediateExecutionContext* ctx, DataType dtype, TensorShape shape,
      absl::optional<std::string> name, const char* raw_device_name,
      const std::vector<std::string>& component_devices,
      std::unique_ptr<Variable>* output);

  // The dtype of the underlying variable.
  DataType dtype();

  // The shape of the underlying variable.
  TensorShape shape();

node {
  name: "x"
  op: "Const"
  attr {
    key: "dtype"
    value {
      type: DT_INT32
    }
  }
  attr {
    key: "value"
    value {
      tensor {
        dtype: DT_INT32
        tensor_shape {
          dim {
            size: 1
          }
        }
        int_val: 1
      }
    }
  }
}

node {
  name: "Placeholder"
  op: "Placeholder"
  attr {
    key: "dtype"
    value {
      type: DT_FLOAT
    }
  }
  attr {
    key: "shape"
    value {
      shape {
        dim {
          size: 2
        }
        dim {
          size: 5
        }
        dim {

  typedef internal::Types$i<$for j, [[T$j]]> type;
};

]]

namespace internal {

# define GTEST_TEMPLATE_ template <typename T> class

// The template "selector" struct TemplateSel<Tmpl> is used to
// represent Tmpl, which must be a class template with one type
// parameter, as a type.  TemplateSel<Tmpl>::Bind<T>::type is defined
// as the type Tmpl<T>.  This allows us to actually instantiate the