%0 = tfl.sub %arg0, %arg1 {fused_activation_function = "RELU6"} : tensor<? x i32>
  func.return %0#0 : tensor<? x i32>
}

// CHECK-LABEL: testSubInt64
func.func @testSubInt64(tensor<? x i64>, tensor<? x i64>) -> tensor<? x i64> {
^bb0(%arg0: tensor<? x i64>, %arg1: tensor<? x i64>):
  // CHECK: tfl.sub %arg0, %arg1 {fused_activation_function = "RELU6"}

  }
  func.func @partitioned_called_func(%arg0: tensor<?xi32>) -> (tensor<?xi32>) {
    func.return %arg0 : tensor<?xi32>
  }
  func.func @stateful_partitioned_call_func(%arg0: tensor<?xi32>) -> (tensor<?xi32>) {
    func.return %arg0 : tensor<?xi32>
  }
  func.func @tpu_partitioned_call_func(%arg0: tensor<?xi32>) -> (tensor<?xi32>) {
    func.return %arg0 : tensor<?xi32>
  }

func.func @argmin(%arg0: tensor<3xi32>, %arg1: tensor<i32>) -> tensor<i32> {
  %0 = "tf.ArgMin"(%arg0, %arg1) : (tensor<3xi32>, tensor<i32>) -> tensor<i32>
  func.return %0 : tensor<i32>

// CHECK-LABEL: argmin
// CHECK:  "tfl.arg_min"(%arg0, %arg1) : (tensor<3xi32>, tensor<i32>) -> tensor<i32>
}

func.func @sigmoid(%arg0: tensor<?x88xf32>) -> tensor<?x88xf32> {
  %0 = "tf.Sigmoid"(%arg0) : (tensor<?x88xf32>) -> tensor<?x88xf32>

  %1 = "tf.Concat"(%0, %arg0, %arg1) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<i32>, tensor<2x1xi32>, tensor<2x1xi32>) -> tensor<2x2xi32>
  func.return %1 : tensor<2x2xi32>
}

// CHECK-LABEL: testConcatCwiseUnary
func.func @testConcatCwiseUnary(%arg0: tensor<?x1xf32>, %arg1: tensor<?x1xf32>, %arg2: tensor<i32>) -> tensor<?x2xf32> {

  func.return %0 : tensor<3xi64>
}

// -----

// CHECK-LABEL: func @testReshape(%arg0: tensor<*xf32>, %arg1: tensor<*xf32>, %arg2: tensor<10000xf32>, %arg3: tensor<*xi32>)
func.func @testReshape(%arg0: tensor<*xf32>, %arg1: tensor<*xf32>, %arg2: tensor<10000xf32>, %arg3: tensor<*xi32>) -> (tensor<100x100xf32>, tensor<*xf32>, tensor<100x100xf32>, tensor<100x100xf32>, tensor<*xf32>, tensor<*xf32>) {

		// unsigned arg0 >> arg2, shifting in bits from arg1 (==(arg1<<64+arg0)>>arg2, keeping low 64 bits), shift amount is mod 64
		{name: "SHRDQ", argLength: 3, reg: gp31shift, asm: "SHRQ", resultInArg0: true, clobberFlags: true},
		// unsigned arg0 << arg2, shifting in bits from arg1 (==(arg0<<64+arg1)<<arg2, keeping high 64 bits), shift amount is mod 64

    // CHECK: tf_device.replicate
    // CHECK-SAME: [%[[ARG_0]], %[[ARG_1]]] as %[[RI_0:[a-z0-9]*]]: tensor<8xi32>
    // CHECK-SAME: [%[[ARG_2]], %[[ARG_3]]] as %[[RI_1:[a-z0-9]*]]: tensor<*xi1>
    // CHECK-SAME: [%[[ARG_4]], %[[ARG_5]]] as %[[RI_2:[a-z0-9]*]]: tensor<*xi32>
    %0:2 = tf_device.replicate([%arg0, %arg1] as %ri: tensor<8xi32>, [%arg2, %arg3] as %ri2: tensor<*xi1>, [%arg4, %arg5] as %ri3: tensor<*xi32>) {n = 2 : i32} {

      ```mlir
        func.func @remove_first_result(%arg0, %arg1) {
          %0:2 = "tf.WhileRegion"(%arg0, %arg1) ({
          ^bb0(%arg2, %arg3):
            %1 = "tf.OpA"() {is_stateless = true}
            "tf.Yield"(%1)
          }, {
          ^bb0(%arg2, %arg3):
            %1 = "tf.OpB"(%arg2) {is_stateless = true}
            %2 = "tf.OpC"(%arg3) {is_stateless = true}
            "tf.Yield"(%1, %2)

// -----

module {
func.func @inference_can_fuse(%arg0: tensor<?x8x8xf32>, %arg1: tensor<8x10xf32>, %arg2: tensor<8x10xf32>, %arg3: tensor<8x40xf32>, %arg4: tensor<10x40xf32>, %arg5: tensor<40xf32>) {
  %0 = "tf.Const"() {_output_shapes = ["tfshape$"], device = "", dtype = f32, value = dense<0.000000e+00> : tensor<f32>} : () -> tensor<f32>

}
// CHECK-LABEL: select
// CHECK-SAME: %[[ARG0:.+]]: tensor<1x3xi1>, %[[ARG1:.+]]: tensor<1x3x!quant.uniform<i8:f32, 2.000000e+00:-1>>, %[[ARG2:.+]]: tensor<1x3x!quant.uniform<i8:f32, 2.000000e+00:-1>>
// CHECK-NOT: stablehlo.select