func.func @fakeQuantPerChannelForActivation(%arg0: tensor<8x4xf32>) -> (tensor<8x4xf32>) {
  %arg1 = arith.constant dense<[0.0, -1.0, 1.0, 0.0]> : tensor<4xf32>
  %arg2 = arith.constant dense<[15.0, 14.0, 16.0, 1.0e-9]> : tensor<4xf32>
  %0 = "tf.FakeQuantWithMinMaxVarsPerChannel"(%arg0, %arg1, %arg2) {num_bits = 3, narrow_range = false} : (tensor<8x4xf32>, tensor<4xf32>, tensor<4xf32>) -> tensor<8x4xf32>
  func.return %0 : tensor<8x4xf32>

// 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": [
        {
          "shape": [1, 4],

  %1 = "tfl.add"(%cst_2, %cst_1) {fused_activation_function = "NONE"} : (tensor<2x2x2xf32>, tensor<  2x2xf32>) -> tensor<2x2x2xf32>
  %2 = "tfl.add"(%cst_0, %cst_2) {fused_activation_function = "NONE"} : (tensor<    2xf32>, tensor<2x2x2xf32>) -> tensor<2x2x2xf32>

  func.return %0, %1, %2 : tensor<2x2xf32>, tensor<2x2x2xf32>, tensor<2x2x2xf32>

}

func.func @main2(%arg0: tensor<2x4xf32>, %arg1: tensor<2x4xf32>) -> tensor<2x4xf32> {
  %0 = "tfl.quantize"(%arg0) {qtype = tensor<2x4x!quant.uniform<u8:f32, 0.49803921568627452>>} : (tensor<2x4xf32>) -> tensor<2x4x!quant.uniform<u8:f32, 0.49803921568627452>>
  %1 = "tfl.quantize"(%arg1) {qtype = tensor<2x4x!quant.uniform<u8:f32, 0.49803921568627452>>} : (tensor<2x4xf32>) -> tensor<2x4x!quant.uniform<u8:f32, 0.49803921568627452>>

  func.return %0 : tensor<3x4x4xf32>
}

func.func @batchmatmulv2_lhs_batch(%arg0: tensor<3x4x2xf32>, %arg1: tensor<2x4xf32>) -> tensor<3x4x4xf32> {
// CHECK-LABEL:   func @batchmatmulv2_lhs_batch
// CHECK:           "mhlo.dynamic_broadcast_in_dim"({{.*}}, {{.*}}) <{broadcast_dimensions = dense<[1, 2]> : tensor<2xi64>}>

  // CHECK: %[[RES0:.*]] = "tfl.batch_matmul"(%arg1, %arg0) <{adj_x = false, adj_y = false, asymmetric_quantize_inputs = false}> : (tensor<8x1024xf32>, tensor<1024x4xf32>) -> tensor<8x4xf32>
  %1 = "tfl.fully_connected"(%0, %arg1, %cst_1) {asymmetric_quantize_inputs = false, fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<4x1024xf32>, tensor<8x1024xf32>, none) -> tensor<4x8xf32>

  // CHECK: return %[[RESULT]]

  %indices = "tf.Const"() {value = dense<[1, 0]> : tensor<2xi32>} : () -> tensor<2xi32>
  %0 = "tf.DynamicStitch"(%indices, %arg0) : (tensor<2xi32>, tensor<2x2xf32>) -> tensor<2x2xf32>
  func.return %0 : tensor<2x2xf32>
}

// CHECK-LABEL: DynamicStitch_scalar_matrix_indices

  %cst_0 = "tf.Const"() {value = dense<0.400000e+00> : tensor<2x2xf32>} : () -> tensor<2x2xf32>
  %0 = "tf.Conv2D"(%arg0, %cst) {data_format = "NHWC", dilations = [1, 1, 2, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x2x2xf32>
  %1 = "tf.Mul"(%0, %cst_0) : (tensor<1x3x2x2xf32>, tensor<2x2xf32>) -> tensor<1x3x2x2xf32>
  func.return %1 : tensor<1x3x2x2xf32>
}

  %0 = stablehlo.uniform_dequantize %arg : (tensor<2x2x!quant.uniform<i8:f32, 1.000000e+0:8>>) -> tensor<2x2xf32>
  return %0 : tensor<2x2xf32>
}
// CHECK-LABEL: uniform_dequantize_op
// CHECK: %[[DEQUANT:.+]] = "tfl.dequantize"({{.*}}) : (tensor<2x2x!quant.uniform<i8:f32, 1.000000e+00:8>>) -> tensor<2x2xf32>
// CHECK: return %[[DEQUANT]]

// -----

  %4 = "tf.MatMul"(%arg0, %3) {device = "", transpose_a = false, transpose_b = false} : (tensor<2x3xf32>, tensor<3x4xf32>) -> tensor<2x4xf32>
  %5 = "tf.Identity"(%4) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
  %6 = "tf.Identity"(%5) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
  func.return %6 : tensor<2x4xf32>

  // CHECK-LABEL: QuantDequantTranspose