/*input_shapes*/ {make_shape({3, 2}), make_shape({2, 4})},
                    /*input_tensors*/ {},
                    /*expected_shape*/ make_shape({3, 4}));

  // Infer shape when second operand has unknown shape.
  CheckOutputShapes(matmul_op,
                    /*input_shapes*/ {make_shape({3, 2}), unknown_shape()},
                    /*input_tensors*/ {},

// be attached. It is intended for use with TFE_OpGetFlatInput to inspect an
// already-finalized operation.
//
// Note that TFE_OpGetFlatInputCount and TFE_OpGetFlatInput operate on a flat
// sequence of inputs, unlike TFE_OpGetInputLength (for getting the length of a
// particular named input list, which may only be part of the op's inputs).

    "true" or "1", this environment variable makes `tf.nn.bias_add` operate
    deterministically (i.e. reproducibly), but currently only when XLA JIT
    compilation is *not* enabled. Setting `TF_DETERMINISTIC_OPS` to "true" or
    "1" also makes cuDNN convolution and max-pooling operate deterministically.
    This makes Keras Conv\*D and MaxPool\*D layers operate deterministically in

/// implements. Each of them is in a 1-to-1 correspondence with the wrapper
/// structures from the first section: these tables only contain function
/// pointers that operate on the corresponding data. Thus, the first argument of
/// each of these functions is a pointer to the paired struct and this argument
/// can be used to track state in between calls (from an object oriented point

struct TF_VariableInfo {
  TF_VariableInfo() = delete;
  // TF_VariableInfo is constructed here by TensorFlow, and will be passed to
  // plugin as a opaque pointer. Plugin will need to call C APIs below to
  // operate on TF_VariableInfo (such as allocate temp tensor for the `var` held
  // by the underlying tensorflow::VariableInfo.
  TF_VariableInfo(int index, const std::string& name, tensorflow::Var* var) {