//
//	':9000' and 'http://localhost:9000/' will return true
func sameLocalAddrs(addr1, addr2 string) (bool, error) {
	// Extract host & port from given parameters
	host1, port1, err := extractHostPort(addr1)
	if err != nil {
		return false, err
	}
	host2, port2, err := extractHostPort(addr2)
	if err != nil {
		return false, err
	}

	var addr1Local, addr2Local bool

	if host1 == "" {

Status Conj(AbstractContext* ctx, AbstractTensorHandle* const input,
            AbstractTensorHandle** output, const char* name = nullptr,
            const char* raw_device_name = nullptr);

// Returns x + y element-wise.
Status AddV2(AbstractContext* ctx, AbstractTensorHandle* const x,
             AbstractTensorHandle* const y, AbstractTensorHandle** z,
             const char* name = nullptr, const char* raw_device_name = nullptr);

//
// is lowered to:
//
//   %sum0 = "tf.AddV2"(%0, %1)
//   %result = "tf.AddV2"(%sum0, %2)
//
// While
//
//   %result = "tf.AddN"(%0, %1, %2, %3, %4)
//
// is lowered to:
//
//   %sum0 = "tf.AddV2"(%0, %1)
//   %sum1 = "tf.AddV2"(%2, %3)
//   %sum2 = "tf.AddV2"(%sum0, %sum1)
//   %result = "tf.AddV2"(%sum2, %4)
//
class LowerAddNOp : public RewritePattern {
 public:

  %cst = arith.constant dense<0.0> : tensor<2x2xf32>
  %0 = "tf.AddV2"(%arg0, %cst) : (tensor<1x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
  func.return %0 : tensor<2x2xf32>

  // CHECK-LABEL: DontRemoveTrivialAdd
  // CHECK: %[[CONST:.*]] = arith.constant dense<0.000000e+00> : tensor<2x2xf32>
  // CHECK: %[[RESULT:.*]] = "tf.AddV2"(%arg0, %[[CONST]]) : (tensor<1x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>

#include "tensorflow/c/experimental/ops/array_ops.h"
#include "tensorflow/c/experimental/ops/math_ops.h"
#include "tensorflow/c/experimental/ops/nn_ops.h"

using std::vector;
using tensorflow::ops::AddV2;
using tensorflow::ops::Div;
using tensorflow::ops::DivNoNan;
using tensorflow::ops::MatMul;
using tensorflow::ops::Mul;
using tensorflow::ops::Neg;
using tensorflow::ops::OnesLike;
using tensorflow::ops::SqrtGrad;

  auto rhs_attr = mlir::dyn_cast_or_null<DenseElementsAttr>(operands[1]);
  if (!rhs_attr && !(is_commutative && lhs_attr)) return {};

  // Mul and Div ops have identity value one while AddV2 and SubOp have identity
  // value zero.
  const int identity =
      (std::is_same<OpT, MulOp>::value || std::is_same<OpT, DivOp>::value ||
       std::is_same<OpT, RealDivOp>::value)
          ? 1
          : 0;

int64_t InferLookupTableFindV2Cost(const CostContext& context,
                                   mlir::TF::LookupTableFindV2Op op) {
  // tf.LookupTableFindV2 ops are usually more costly than tf.AddV2 with the
  // same input size, as it involves more operations like hashing, map lookup,
  // etc.
  constexpr int64_t kLookupTableFindCostScale = 8;
  constexpr int64_t kLookupTableFindStringKeyCostScale = 16;

        %1 = "tf.Const"() {_xla_outside_compilation = "0", value = dense<1.0> : tensor<f32>} : () -> (tensor<f32>)
        %2 = "tf.Identity"(%1) {_xla_outside_compilation = "0"} : (tensor<f32>) -> (tensor<f32>)
        %3 = "tf.AddV2"(%1, %2) : (tensor<f32>, tensor<f32>) -> (tensor<f32>)
        tf_device.return %3 : tensor<f32>
      }) {num_cores_per_replica = 1, topology =  "", device_assignment =  []} : () -> tensor<f32>
      return %0 : tensor<f32>

    num_args: The number of inputs.
    device: The tensorflow device. eg. "/CPU:0"
    op_attrs: The tensorflow attributes excluding the func attrs.
    op_name: The tensorflow op name. eg. "tf.AddV2"

    Example:
      tfrt_fallback_sync.createop() key(0) device("/CPU:0")
        "some.op"() {attr1 = value, attr2 = value2} {f_attr1 = "fn1", f_attr2 = "fn2"} num_args(1)
  }];

  let arguments = (ins

  %28 = "tf.AddV2"(%27, %3) {device = ""} : (tensor<2x16xi32>, tensor<i32>) -> tensor<2x16xi32>
  %29 = "tf.AddV2"(%17, %28) {device = ""} : (tensor<2x16xi32>, tensor<2x16xi32>) -> tensor<2x16xi32>
  %30 = "tf.GatherV2"(%9, %29, %5) {batch_dims = 0 : i64, device = ""} : (tensor<32x1xf32>, tensor<2x16xi32>, tensor<i32>) -> tensor<2x16x1xf32>