func.func @main(%arg0: tensor<1x4xf32>) -> tensor<1x3xf32> {
  %0 = stablehlo.constant dense<1.0> : tensor<4x3xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
  return %1 : tensor<1x3xf32>
}
// CHECK: @main(%[[ARG_0:.+]]: tensor<1x4xf32>) -> tensor<1x3xf32>

  %43 = "quantfork.stats"(%42) {layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>} : (tensor<1x4x3xf32>) -> tensor<1x4x3xf32>
  %44 = "tfl.strided_slice"(%43, %1, %2, %3) {begin_mask = 7 : i32, ellipsis_mask = 0 : i32, end_mask = 5 : i32, new_axis_mask = 0 : i32, shrink_axis_mask = 0 : i32, offset = false} : (tensor<1x4x3xf32>, tensor<3xi32>, tensor<3xi32>, tensor<3xi32>) -> tensor<1x2x3xf32>

// -----

// 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>

  %0:3 = "tf.Split"(%arg0, %arg1) : (tensor<i32>, tensor<1x4x3x3xf32>) -> (tensor<1x4x3xf32>, tensor<1x4x3xf32>, tensor<1x4x3xf32>)
  func.return %0#0 : tensor<1x4x3xf32>

  // CHECK-LABEL: split
  // CHECK: "tfl.split"(%arg0, %arg1) <{num_splits = 3 : i32}> : (tensor<i32>, tensor<1x4x3x3xf32>) -> (tensor<1x4x3xf32>, tensor<1x4x3xf32>, tensor<1x4x3xf32>)
}

// CHECK: %0 = "tfl.pseudo_const"() <{value = dense<2> : tensor<2xi32>}> : () -> tensor<2xi32>
// CHECK: %1 = "tfl.reshape"(%arg0, %0) : (tensor<1x4xf32>, tensor<2xi32>) -> tensor<2x2xf32>

{
  "version": 3,
  "operator_codes": [
    {
      "builtin_code": "RESHAPE"
    }
  ],
  "subgraphs": [
    {
      "tensors": [
        {

    %recurrent_stats = "quantfork.stats"(%recurrent_input) {layerStats = dense<[-2.0, 1.0]> : tensor<2xf32>} : (tensor<1x20xf32>) -> tensor<1x20xf32>
    %cell_input = arith.constant dense<1.0> : tensor<1x20xf32>
    %cell_stats = "quantfork.stats"(%cell_input) {layerStats = dense<[-2.73090601, 7.94872093]> : tensor<2xf32>} : (tensor<1x20xf32>) -> tensor<1x20xf32>
    %0 = "tfl.unidirectional_sequence_lstm"(%arg0,

// CHECK-DAG: %[[input_18:.*]] = "quantfork.stats"({{.*}}) <{layerStats = dense<[-8.000000e-01, 1.600000e+00]> : tensor<2xf32>}> : (tensor<1x4xf32>) -> tensor<1x4xf32>
// CHECK-DAG: %[[input_19:.*]] = "quantfork.stats"({{.*}}) <{layerStats = dense<[-2.000000e+00, 4.000000e+00]> : tensor<2xf32>}> : (tensor<1x2xf32>) -> tensor<1x2xf32>

  %0 = "tf.Cast"(%arg0) : (tensor<!tf_type.resource<tensor<8x40xf32>>>) -> tensor<*x!tf_type.resource>
  %1 = "tf.ReadVariableOp"(%0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<*x!tf_type.resource>) -> tensor<8x40xf32>
  func.return %1: tensor<8x40xf32>

    %recurrent_stats = "quantfork.stats"(%recurrent_input) {layerStats = dense<[-2.0, 1.0]> : tensor<2xf32>} : (tensor<1x20xf32>) -> tensor<1x20xf32>
    %cell_input = arith.constant dense<1.0> : tensor<1x20xf32>
    %cell_stats = "quantfork.stats"(%cell_input) {layerStats = dense<[-2.73090601, 7.94872093]> : tensor<2xf32>} : (tensor<1x20xf32>) -> tensor<1x20xf32>

  %2 = "tf.Add"(%1, %1) {
    _xla_compile_device_type = "TPU", _replication_info = "cluster",
    device = "/task:0/device:TPU:0", dtype = f32
  } : (tensor<1x80xf32>, tensor<1x80xf32>) -> tensor<1x80xf32>
  %3 = "tf.ResourceGatherNd"(%arg0, %0) {
    Tindices = i32
  } : (tensor<*x!tf_type.resource<tensor<80xf32>>>, tensor<i32>) -> tensor<1x80xf32>