func.return %0 : tensor<8x16xf32>

// CHECK-LABEL:minimum
// CHECK:  "tfl.minimum"(%arg0, %arg1) : (tensor<8x16xf32>, tensor<8x16xf32>) -> tensor<8x16xf32>
}

func.func @realDiv(%arg0: tensor<8x16xf32>, %arg1: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.RealDiv"(%arg0, %arg1) : (tensor<8x16xf32>, tensor<8x16xf32>) -> tensor<8x16xf32>

  // CHECK-LABEL: func @parallel_execute_with_tiled_input
  // CHECK-SAME: (%[[ARG_0:[a-z0-9]*]]: tensor<128x10xf32>, %[[ARG_1:[a-z0-9]*]]: tensor<128x10xf32>, %[[ARG_2:[a-z0-9]*]]: tensor<*xi32>, %[[ARG_3:[a-z0-9]*]]: tensor<*xi32>)
  func.func @parallel_execute_with_tiled_input(%arg0: tensor<128x10xf32>, %arg1: tensor<128x10xf32>, %arg2: tensor<*xi32>, %arg3: tensor<*xi32>) -> (tensor<*xi32>, tensor<*xi1>) {
    // CHECK: tf_device.replicate

  // CHECK: return %0
}

// CHECK-LABEL: testAddOfNegRight
func.func @testAddOfNegRight(%arg0: tensor<8x16xf32>, %arg1: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.Neg"(%arg1) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  %1 = "tf.Add"(%arg0, %0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<8x16xf32>, tensor<8x16xf32>) -> tensor<8x16xf32>

  %11 = "tf.Floor"(%10#0) {device = ""} : (tensor<2x16xf32>) -> tensor<2x16xf32>
  %12 = "tf.Maximum"(%0, %11) {device = ""} : (tensor<f32>, tensor<2x16xf32>) -> tensor<2x16xf32>
  %13 = "tf.Minimum"(%12, %4) {device = ""} : (tensor<2x16xf32>, tensor<f32>) -> tensor<2x16xf32>
  %14 = "tf.Cast"(%13) {Truncate = false, device = ""} : (tensor<2x16xf32>) -> tensor<2x16xi32>

  func.return %0 : tensor<8x16xf32>
}

// -----

func.func @testCumprod(%arg: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %axis = arith.constant dense<-3> : tensor<i32>
  // expected-error @+1 {{axis operand should be within range [-2, 2)}}
  %0 = "tf.Cumprod"(%arg, %axis) : (tensor<8x16xf32>, tensor<i32>) -> tensor<8x16xf32>
  func.return %0 : tensor<8x16xf32>
}

    }) {device = "/device:CPU:0"} : () -> tensor<*xf32>
    // CHECK: "tf.Cast"(%{{.*}}) <{Truncate = false}> : (tensor<1x8x2xf32>) -> tensor<*xf32>
    // CHECK: (tensor<i32>, tensor<1x8x2xf32>) -> (tensor<1x8x1xf32>, tensor<1x8x1xf32>)
    %3:2 = "tf.Split"(%0, %1) {device = ""} : (tensor<i32>, tensor<*xf32>) -> (tensor<*xf32>, tensor<*xf32>)
    %4 = tensor.cast %1 : tensor<*xf32> to tensor<?x?x?xf32>

// dimensions into a single dimension. For example,
//
//   %shape = arith.constant dense<[1, 128, 64]> : tensor<3xi32>
//   %reshape = tfl.reshape(%input, %shape) // %input: tensor<128x64xf32>
//   %fc = tfl.fully_connected(%reshape, %filter, %bias)
//           {keep_num_dims = false, weights_format = "DEFAULT"}
//
// can be canonicalized to
//