%0 = "tf.Cast"(%arg0) {Truncate = false, device = ""} : (tensor<1x1x10xf32>) -> tensor<1x1x10xbf16>
  %1 = "tf.BatchMatMulV2"(%0, %cst) {adj_x = false, adj_y = false, device = ""} : (tensor<1x1x10xbf16>, tensor<10x2xbf16>) -> tensor<1x1x2xbf16>
  %2 = "tf.Cast"(%1) {Truncate = false} : (tensor<1x1x2xbf16>) -> tensor<1x1x2xf32>
  %3 = "tf.IdentityN"(%2) {device = ""} : (tensor<1x1x2xf32>) -> tensor<1x1x2xf32>

  %8 = mhlo.and %5, %7 : tensor<10x10xi1>
  %9 = mhlo.subtract %arg0, %4 : tensor<10x10xbf16>
  %10 = mhlo.divide %9, %1 : tensor<10x10xbf16>
  %11 = mhlo.add %10, %3 : tensor<10x10xbf16>
  %12 = "mhlo.select"(%8, %11, %10) : (tensor<10x10xi1>, tensor<10x10xbf16>, tensor<10x10xbf16>) -> tensor<10x10xbf16>
  %13 = "mhlo.round_nearest_afz"(%12) : (tensor<10x10xbf16>) -> tensor<10x10xbf16>

  static constexpr char kModuleWithNonConstParam[] = R"(
  module attributes {tf.versions = {bad_consumers = [], min_consumer = 0 : i32, producer = 1610 : i32}} {
    func.func @main(%arg0: tensor<3x3x10xbf16>, %arg1: tensor<3xi32>) -> tensor<1x?x4xbf16> attributes {allow_soft_placement = false, tf.entry_function = {control_outputs = "", inputs = "_arg0,_arg1,_arg2", outputs = "_retval0"}} {

    return %values : tensor<10x10xbf16>
  }

  // CHECK-LABEL: fusedBatchNormV3_noTraining