%3 = "tfl.dequantize"(%2) : (tensor<1x3x3x!quant.uniform<i8:f32, 0.004:-128>>) -> tensor<1x3x3xf32>
  %4 = "tfl.div"(%arg0, %3) {fused_activation_function = "NONE"} : (tensor<1x3x3xf32>, tensor<1x3x3xf32>) -> tensor<1x3x3xf32>
  func.return %4 : tensor<1x3x3xf32>
//  CHECK: %[[logistic:.*]] = "tfl.logistic"
//  CHECK: %[[dq:.*]] = "tfl.dequantize"(%[[logistic]])

  // CHECK: [[Result:%.*]] = "tf.Case"(%arg0, %arg1)
  // CHECK-SAME: branches = [@tf.CaseRegion_branch0{{.*}}, @tf.CaseRegion_branch1{{.*}}]
  // CHECK-SAME: is_stateless = false
  %0 = "tf.CaseRegion"(%arg0) ({
    %1 = "tf.Cast"(%arg1) {Truncate = false} : (tensor<!tf_type.resource<tensor<1x2x3xf32>>>) -> tensor<!tf_type.resource>

}

// CHECK-LABEL: test_gather
func.func @test_gather(%arg0 : tensor<1x2x20xf32>, %arg1 : tensor<3x5xi32>) -> (tensor<1x3x5x20xf32>){
    // expected-remark@+1 {{Cost: 1}}
    %0 = "tf.Const"() { value = dense<[1]> : tensor<1xi32> } : () -> tensor<1xi32>
    // expected-remark@+1 {{Cost: 300}}
    %1 = "tf.GatherV2"(%arg0, %arg1, %0) : (tensor<1x2x20xf32>, tensor<3x5xi32>, tensor<1xi32>) -> tensor<1x3x5x20xf32>

    %2 = "quantfork.stats"(%1) {layerStats = dense<[5.00000000e-6, 7.00000000e-1]> : tensor<2xf32>} : (tensor<2x2x3xf32>) -> tensor<2x2x3xf32>
    return %2 : tensor<2x2x3xf32>
  }
// CHECK: func.func private @quantize_dot_general_batch_per_tensor_quantized_fn(%[[ARG_0:.+]]: tensor<2x2x2xf32>) -> tensor<2x2x3xf32> attributes {tf._original_func_name = "main_0"}

// CHECK: [[UNPACK:%.*]]:3 = "tf.Unpack"(%arg0) <{axis = 0 : i64}> : (tensor<3x2x2xf32>) -> (tensor<2x2xf32>, tensor<2x2xf32>, tensor<2x2xf32>)
// CHECK: [[SCALAR_ZERO:%.*]] = arith.constant dense<0> : tensor<i32>
// CHECK: [[CONCAT:%.*]] = "tf.Concat"([[SCALAR_ZERO]], [[UNPACK]]#0, [[UNPACK]]#1, [[UNPACK]]#2) : (tensor<i32>, tensor<2x2xf32>, tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<?x2xf32>
// CHECK: [[LENGTHS:%.*]] = arith.constant dense<0> : tensor<0xi64>

  // CHECK: %[[v1:.*]] = "tf.BatchMatMulV2"(%arg0, %[[v0]]) <{adj_x = false, adj_y = false}> : (tensor<5x4x3xf32>, tensor<1x3x2xf32>) -> tensor<5x4x2xf32>
  // CHECK: %[[v2:.*]] = "tf.Transpose"(%[[v1]], %[[cst_1]]) : (tensor<5x4x2xf32>, tensor<3xi32>) -> tensor<4x2x5xf32>
  // CHECK: return %[[v2]] : tensor<4x2x5xf32>

// CHECK-NEXT:    return %2 : tensor<3x2xf32>
}

func.func @transpose_3d(%arg0: tensor<1x2x3xf32>) -> tensor<3x2x1xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[2, 1, 0]> : tensor<3xi64>}> : (tensor<1x2x3xf32>) -> tensor<3x2x1xf32>
  func.return %0 : tensor<3x2x1xf32>

// CHECK-LABEL:   transpose_3d

  func.func @tf_mod(%arg1: tensor<2x2xf32>) -> tensor<2x2xf32> {
    %cst = "tf.Const"() {value = dense<7.000000e+00> : tensor<f32>} : () -> tensor<f32>
    // CHECK: "mhlo.dynamic_broadcast_in_dim"
    // CHECK: mhlo.remainder
    %6 = "tf.Mod"(%arg1, %cst) {_global_shape = [#tf_type.shape<4x8>], device = ""} : (tensor<2x2xf32>, tensor<f32>) -> tensor<2x2xf32>
    return %6 : tensor<2x2xf32>
  }

  // CHECK-LABEL: func @concat_v2

  %1 = "tf.TensorListGetItem"(%0, %arg2, %arg1) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<!tf_type.variant<tensor<1x32xf32>>>, tensor<i32>, tensor<2xi32>) -> tensor<1x32xf32>
  %2 = "tf.TensorListGetItem"(%0, %arg3, %arg1) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<!tf_type.variant<tensor<1x32xf32>>>, tensor<i32>, tensor<2xi32>) -> tensor<1x32xf32>

//   %arg1)
//     : (tensor<i32>, tensor<2x2xi32>, tensor<2xf32>, tensor<2x2x2xf32>)
//     -> tensor<5x2xf32>
//
// is lowered to
//
//   %shape = "tf.Const"() {value = dense<[-1, 2]> : tensor<2xi64>}
//   %inp0 = "tf.Reshape"(%arg0, %shape)
//     : (tensor<2xf32>, tensor<2xi64>) -> tensor<1x2xf32>
//   %inp1 = "tf.Reshape"(%arg1, %shape)
//     : (tensor<2x2x2xf32>, tensor<2xi64>) -> tensor<4x2xf32>