func.func @uniform_quantize_and_dequantize_per_axis(%arg0 : tensor<2x2xf32>) -> tensor<2x2xf32> {
  %scales = "tf.Const"() { value = dense<[1.0, 2.0]> : tensor<2xf32> } : () -> tensor<2xf32>
  %zps = "tf.Const"() { value = dense<[3, 4]> : tensor<2xi32> } : () -> tensor<2xi32>

  // CHECK: %[[QUANTIZE:.*]] = mhlo.uniform_quantize %arg0 : (tensor<2x2xf32>) -> tensor<2x2x!quant.uniform<i8:f32:0, {1.000000e+00:3,2.000000e+00:4}>>

  %5 = "tf.TPUExecute"(%3#0, %1, %program#0) {_producer_name = "UNKNOWN", device = "/TPU:0"} : (tensor<1x1xf32>, tensor<1x1xf32>, tensor<3x!tf_type.string>) -> tensor<1x1xf32>
  %6 = "tf.TPUExecute"(%3#1, %1, %program#1) {_producer_name = "UNKNOWN", device = "/TPU:1"} : (tensor<1x1xf32>, tensor<1x1xf32>, tensor<3x!tf_type.string>) -> tensor<1x1xf32>

func.func @pack(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<2x1xf32> {
  %0 = "tfl.pack"(%arg0, %arg1) {axis = 0 : i32, values_count = 2 : i32} : (tensor<1xf32>, tensor<1xf32>) -> tensor<2x1xf32>
  func.return %0 : tensor<2x1xf32>
}

// CHECK:   func @pack(%[[VAL_0:.*]]: tensor<1xf32>, %[[VAL_1:.*]]: tensor<1xf32>) -> tensor<2x1xf32> {
// CHECK-DAG:       %[[VAL_2:.*]] = "tfl.pseudo_const"{{.*}}dense<1> : tensor<4xi32>

// CHECK: %[[v2:.*]] = "tf.Reshape"(%[[v1]], %[[cst_1]]) : (tensor<2x1x2xf32>, tensor<4xi64>) -> tensor<2x1x1x2xf32>
// CHECK: return %[[v2]] : tensor<2x1x1x2xf32>
}

func.func @einsum_with_runtime_outputshape1(%arg0 : tensor<?x36x32xf32>, %arg1 : tensor<?x36x?x32xf32>) -> tensor<?x36x?xf32> {

    %4:4 = "tf.CustomAggregator"(%3) {calibration_method = 1 : i32, id = "1", max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32, num_bins = 0 : i32} : (tensor<3x11xf32>) -> (tensor<3x11xf32>, tensor<f32>, tensor<f32>, tensor<*xi64>)
    %5 = stablehlo.add %4#0, %1 : tensor<3x11xf32>
    %6 = stablehlo.multiply %5, %1 : tensor<3x11xf32>

func.func @RemoveRedundantUnpackPack(%arg0: tensor<2x5xf32>) -> tensor<2x5xf32> {
  %0:2 = "tfl.unpack"(%arg0) {axis = 0 : i32, num = 2 : i32} : (tensor<2x5xf32>) -> (tensor<5xf32>, tensor<5xf32>)
  %1 = "tfl.pack"(%0#0, %0#1) {axis = 0 : i32, values_count = 2 : i32} : (tensor<5xf32>, tensor<5xf32>) -> (tensor<2x5xf32>)
  func.return %1: tensor<2x5xf32>
  // CHECK-NOT: pack
  // CHECK: return %arg0 : tensor<2x5xf32>
}

// -----

  // CHECK-LABEL: @xla_svd
  func.func @xla_svd(%arg0: tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>) {
    // CHECK-NOT: XlaSvd
    %s, %u, %v = "tf.XlaSvd"(%arg0) {max_iter = 1, epsilon = 1.0E-09 : f32, precision_config = ""} : (tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>)
    func.return %s, %u, %v : tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>
  }

  func.func @identity(%arg0: f32) -> f32 {

}
func.func @_func(%arg0: tensor<2x4xf32>, %arg1: tensor<4x2xf32>) -> tensor<2x2xf32> {
  %0 = "tf.MatMul"(%arg0, %arg1) {_XlaSharding = "\08\03\1A\02\02\01\22\02\00\01"} : (tensor<2x4xf32>, tensor<4x2xf32>) -> tensor<2x2xf32>
  %1 = "tf.Identity"(%0) : (tensor<2x2xf32>) -> tensor<2x2xf32>
  return %1 : tensor<2x2xf32>
}

// -----
// The following op sharding is used in the following test case:

// -----

// CHECK-LABEL: float_matmul
func.func @float_matmul(
  %arg0: tensor<1x10xf32>, %arg1: tensor<10x10xf32>) -> (tensor<*xf32>, tensor<*xf32>, tensor<*xf32>) {
  %cst = "tf.Const"() {value = dense<0.000000e+00> : tensor<10xf32>} : () -> tensor<10xf32>
  %0 = "tf.MatMul"(%arg0, %arg1) {
    transpose_a = false, transpose_b = false
  } : (tensor<1x10xf32>, tensor<10x10xf32>) -> tensor<*xf32>

// CHECK-LABEL: main
// CHECK: return %arg0 : tensor<5x7xf32>
}

// - transpose
//
func.func @transpose_2d(%arg0: tensor<2x3xf32>) -> tensor<3x2xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[1, 0]> : tensor<2xi64>}> : (tensor<2x3xf32>) -> tensor<3x2xf32>
  func.return %0 : tensor<3x2xf32>

// CHECK-LABEL:   transpose_2d
// CHECK-NEXT:    %0 = "tfl.pseudo_const"() <{value = dense<[1, 0]> : tensor<2xi64>}> : () -> tensor<2xi64>