tensor<1x1x5xf32>,
        tensor<2x5xf32>, tensor<2x5xf32>, tensor<2x5xf32>, tensor<2x5xf32>,
        tensor<2x4xf32>, tensor<2x4xf32>, tensor<2x4xf32>, tensor<2x4xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<4x2xf32>, tensor<4xf32>,
        tensor<1x4xf32>, tensor<1x2xf32>,

  // CHECK:  %[[MUL1:.*]] = "tf.Mul"(%arg1, %[[RSQRT]]) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
  // CHECK:  %[[MUL2:.*]] = "tf.Mul"(%arg0, %[[MUL1]]) : (tensor<1x1x6x2xf32>, tensor<2xf32>) -> tensor<1x1x6x2xf32>
  // CHECK:  %[[MUL3:.*]] = "tf.Mul"(%[[MEAN]], %[[MUL1]]) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
  // CHECK:  %[[SUB:.*]] = "tf.Sub"(%arg2, %[[MUL3]]) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>

    %0 = "tf.MatMul"(%arg0, %arg1) {attr_map = "0:transpose_a,1:transpose_b", device = "", transpose_a = false, transpose_b = false} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
    return %0 : tensor<2x2xf32>
  }

func.func @BiasAddGrad_unranked(%arg0: tensor<*xf32>) -> tensor<?xf32> {
  // CHECK: tf.BiasAddGrad
  %0 = "tf.BiasAddGrad"(%arg0) {data_format = "NCHW"} : (tensor<*xf32>) -> tensor<?xf32>
  func.return %0 : tensor<?xf32>
}

// CHECK-LABEL: func @rsqrt_grad
// CHECK-SAME: (%[[ARG0:.*]]: tensor<2xf32>, %[[ARG1:.*]]: tensor<2xf32>)
func.func @rsqrt_grad(%arg0: tensor<2xf32>, %arg1: tensor<2xf32>) -> tensor<2xf32> {

    %2 = "quantfork.stats"(%1) {layerStats = dense<[5.00000000e-6, 7.00000000e-1]> : tensor<2xf32>} : (tensor<?x3xf32>) -> tensor<?x3xf32>
    return %2 : tensor<?x3xf32>
  }
// CHECK: func.func private @quantize_dot_general_with_bias_dynamic_fn(%[[ARG_0:.+]]: tensor<?x2xf32>) -> tensor<?x3xf32> attributes {tf._original_func_name = "main_0"}

//     : (tensor<1x2xf32>) -> tensor<2xf32>
//   %items1:4 = "tf.Unpack"(%[[INP1]]) {axis = 0 : i64}
//     : (tensor<4x2xf32>) -> (tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
//     tensor<2xf32>)
//   %axis = "tf.Const"() {value = dense<0> : tensor<i64>}
//   %0 = "tf.Pack"(items1#3, items1#2, items1#1, items1#0, %items0, %axis)
//     : (tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,

  %4 = "tfl.minimum"(%0, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %5 = "tfl.minimum"(%1, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %6 = "tfl.minimum"(%2, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>
  %7 = "tfl.minimum"(%3, %cst) : (tensor<32xf32>, tensor<32xf32>) -> tensor<32xf32>

  %5 = shape.shape_of %4 : tensor<?x10xf32> -> tensor<2xindex>
  %6 = stablehlo.dynamic_broadcast_in_dim %1, %5, dims = [1] : (tensor<10xf32>, tensor<2xindex>) -> tensor<?x10xf32>
  %7 = stablehlo.add %4, %6 : tensor<?x10xf32>
  %8 = stablehlo.clamp %2, %7, %3 : (tensor<f32>, tensor<?x10xf32>, tensor<f32>) -> tensor<?x10xf32>
  func.return %8: tensor<?x10xf32>
}

    %0 = "tf.Div"(%arg1, %cst_0) : (tensor<2x2xf32>, tensor<f32>) -> tensor<2x2xf32>
    %1 = "tf.AddV2"(%0, %cst_1) : (tensor<2x2xf32>, tensor<f32>) -> tensor<2x2xf32>
    %2 = "tf.Floor"(%1) : (tensor<2x2xf32>) -> tensor<2x2xf32>
    %3 = "tf.ClipByValue"(%2, %cst_5, %cst_6) : (tensor<2x2xf32>, tensor<f32>, tensor<f32>) -> tensor<2x2xf32>
    %4 = "tf.Cast"(%3) {Truncate = false} : (tensor<2x2xf32>) -> tensor<2x2xi8>

// CHECK-SAME: (%[[ARG_0:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>, %[[ARG_1:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>, %[[ARG_2:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>, %[[ARG_3:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>, %[[ARG_4:.*]]: tensor<!tf_type.resource<tensor<32xf32>>>)