%0 = "tf.Const"() {value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>

  %2 = "tf.Const"() {value = dense<0.0> : tensor<2x2xf32>} : () -> tensor<2x2xf32>
  %3 = "tf.Const"() {value = dense<0.0> : tensor<4x4xf32>} : () -> tensor<4x4xf32>
  "tf.SendTPUEmbeddingGradients"(%2, %3) {_tpu_embedding_layer = "call1", config = "\0A\0B\0C\0D", operandSegmentSizes = array<i32: 2, 0>} : (tensor<2x2xf32>, tensor<4x4xf32>) -> ()

    %3 = "tf.Mul"(%1, %2) : (tensor<10xf32>, tensor<10xf32>) -> tensor<10xf32>
    return %3 : tensor<10xf32>
  }
}

// -----

// CHECK-LABEL: module
module attributes {tf_saved_model.semantics} {
  // CHECK: ml_program.global{{.*}}mutable{{.*}}@vars.v
  "tf_saved_model.global_tensor"() { is_mutable, sym_name = "v", type = tensor<10xf32>, value = dense<[0.,10.,2.,3.,4.,5.,6.,7.,8.,9.]> : tensor<10xf32> } : () -> ()

module{
  func.func @main(%arg0: tensor<10xf32>) -> tensor<10xf32> {
    return %arg0 : tensor<10xf32>
  }
  func.func @func1(%arg0: tensor<10xf32>, %arg1: tensor<10xf32>) -> tensor<10xf32> {
    %0 = stablehlo.add %arg0, %arg1 : tensor<10xf32>
    %1 = stablehlo.add %arg0, %arg1 : tensor<10xf32>
    return %0 : tensor<10xf32>
  }
})mlir";

  %elem_shape = "tf.Const"() {value = dense<> : tensor<0xi32>} : () -> tensor<0xi32>
  // CHECK-NEXT: %[[BUFFER:.*]] = "tf.Identity"(%[[ARG1]]) : (tensor<10xf32>) -> tensor<10xf32>
  // CHECK-NEXT: %[[SIZE:.*]] = "tf.Const"() <{value = dense<10> : tensor<1xi32>}> : () -> tensor<1xi32>
  %tl = "tf.TensorListFromTensor"(%arg1, %elem_shape) : (tensor<10xf32>, tensor<0xi32>) -> tensor<!tf_type.variant<tensor<f32>>>

    %1 = "tf.BiasAdd"(%0, %arg2) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<2xf32>) -> tensor<*xf32>
    %2 = "tf.Relu6"(%1) : (tensor<*xf32>) -> tensor<*xf32>
    func.return %2 : tensor<*xf32>
  }

// CHECK-LABEL: func @conv_with_single_layer

// CHECK: %[[quantize:.*]] = "tf.PartitionedCall"(%arg0

^bb0(%arg0: tensor<40x37xf32>, %arg1: tensor<40x37xf32>):
  %cst = arith.constant dense<1.0> : tensor<40xf32>
  %0:2 = "tfl.fully_connected"(%arg0, %arg1, %cst) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<40x37xf32>, tensor<40x37xf32>, tensor<40xf32>) -> (tensor<40x40xf32>, tensor<40x40xf32>)
  func.return %0 : tensor<40x40xf32>

    %a:2, %control1 = tf_executor.island wraps "tf.A"() {_class = ["loc:@class"]} : () -> (tensor<2xf32>, tensor<2xf32>)
    %s:2, %control2 = tf_executor.island wraps "tf.Split"(%c, %a#1) {num_split = 2 : i32} : (tensor<i32>, tensor<2xf32>) -> (tensor<1xf32>, tensor<1xf32>)
    tf_executor.fetch
  }
  func.return
}

// -----

// CHECK-LABEL: func @no_colocate_split_has_device

// CHECK-NEXT:  %[[quant:.*]] = "tfl.quantize"(%[[split]]#0) <{qtype = tensor<2x!quant.uniform<u8:f32, 1.000000e+00>>}> {volatile} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.000000e+00>>
// CHECK-NEXT:  return %[[quant]] : tensor<2x!quant.uniform<u8:f32, 1.000000e+00>>

func.func @transpose_simple_1d() -> tensor<2xf32> {
  %0 = stablehlo.constant dense<[0.000000e+0, 1.000000e+0]> : tensor<2xf32>
  %1 = stablehlo.transpose %0, dims = [0] : (tensor<2xf32>) -> tensor<2xf32>
  return %1 : tensor<2xf32>
}
// CHECK-DAG: %[[CONST_0:.+]] = stablehlo.constant dense<[0.000000e+00, 1.000000e+00]> : tensor<2xf32>
// CHECK-NOT: transpose
// CHECK: return %[[CONST_0]] : tensor<2xf32>

// -----

      %1 = "tf.Const"() {value = dense<1.0> : tensor<2xf32>} : () -> tensor<2xf32>
      %2 = "tf.Const"() {value = dense<2> : tensor<2xi32>} : () -> tensor<2xi32>
      %3 = "tf.Const"() {value = dense<1.0> : tensor<2xf32>} : () -> tensor<2xf32>
      %4 = "tf.Const"() {value = dense<0> : tensor<2xi32>} : () -> tensor<2xi32>