%10 = "tf.AddV2"(%9, %cst_2) {device = ""} : (tensor<2x2x1x3xi32>, tensor<3xi32>) -> tensor<2x2x1x3xi32>
    %11 = "tf.Cast"(%10) {Truncate = false, device = ""} : (tensor<2x2x1x3xi32>) -> tensor<2x2x1x3xf32>
    %12 = "tf.Mul"(%11, %cst) {device = ""} : (tensor<2x2x1x3xf32>, tensor<f32>) -> tensor<2x2x1x3xf32>

      tf_executor.fetch %outputs#1, %outputs#0 : tensor<2x2x1x2xi32>, tensor<2x2x1x2xf32>
    }
    func.return %0#0, %0#1 : tensor<2x2x1x2xi32>, tensor<2x2x1x2xf32>
  }
}

// CHECK:        name: "serving_default_x"
// CHECK-NEXT:   op: "_Arg"

    %12 = "quantfork.qcast"(%11) {volatile} : (tensor<2x3x1x3xf32>) -> tensor<2x3x1x3x!quant.uniform<i8:f32, 3.000000e-01:1>>
    %13 = "quantfork.dcast"(%12) : (tensor<2x3x1x3x!quant.uniform<i8:f32, 3.000000e-01:1>>) -> tensor<2x3x1x3xf32>
    return %13 : tensor<2x3x1x3xf32>
  }

  // CHECK: quantized_dot_general_fn_1

  %3 = "tfl.quantize"(%2) {qtype = tensor<1x2x1x3x!quant.uniform<i8:f32, 1.0>>, volatile} : (tensor<1x2x1x3xf32>) -> tensor<1x2x1x3x!quant.uniform<i8:f32, 1.0>>
  %5 = "tfl.dequantize"(%arg0) : (tensor<1x2x1x3x!quant.uniform<i8:f64, 1.0>>) -> tensor<1x2x1x3xf32>
  "tfl.assign_variable"(%1, %5) : (tensor<!tf_type.resource>, tensor<1x2x1x3xf32>) -> ()

  %3 = mhlo.multiply %1, %2 : tensor<1x2x1x4xi32>
  // CHECK:      return %[[RES]] : tensor<1x2x1x4xi32>
  return %3 : tensor<1x2x1x4xi32>
}

// CHECK: @notFoldBroadcastInDimBeforeMulOpWhenArgIsNonConst_bcast_dim_1D_int(%[[ARG:.*]]: tensor<1x1x2x4xi32>) -> tensor<1x1x2x4xi32>
func.func @notFoldBroadcastInDimBeforeMulOpWhenArgIsNonConst_bcast_dim_1D_int(%arg0: tensor<1x1x2x4xi32>) -> (tensor<1x1x2x4xi32>) {

    : (tensor<2x3x2x1xi8>, tensor<2x1x1x1xi8>) -> tensor<2x2x2x1xi32>
  %1 = chlo.broadcast_add %0, %zp_offset : (
      tensor<2x2x2x1xi32>, tensor<2x2x2x1xi32>) -> tensor<2x2x2x1xi32>
  %2 = chlo.broadcast_add %1, %bias : (
      tensor<2x2x2x1xi32>, tensor<1xi32>) ->tensor<2x2x2x1xi32>
  return %2 : tensor<2x2x2x1xi32>
}

// -----

// CHECK-LABEL: func @convolution_add_subtract

      stablehlo.return %2 : tensor<f32>
  }) {padding = dense<[[0, 0], [1, 1], [1, 1], [0, 0]]> : tensor<4x2xi64>, window_dimensions = array<i64: 1, 3, 3, 1>} : (tensor<2x3x1x3xf32>, tensor<f32>) -> tensor<2x3x1x3xf32>
  return %1 : tensor<2x3x1x3xf32>
}

// -----

// CHECK-LABEL: @bitcast_convert_i32_f32(%arg0: tensor<1x256128xi32>) -> tensor<1x256128xbf16>

  %cst = "tf.Const"() {value = dense<[0, 2, 1, 3]> : tensor<4xi32>} : () -> tensor<4xi32>
  %0 = "tf.AddV2"(%arg0, %arg1) {device = ""} : (tensor<1x1x2x3xf32>, tensor<2x3xf32>) -> tensor<1x1x2x3xf32>
  %1 = "tf.Transpose"(%0, %cst) {device = ""} : (tensor<1x1x2x3xf32>, tensor<4xi32>) -> tensor<1x2x1x3xf32>

  func.return %1 : tensor<1x2x1x3xf32>

  %res_p3 = "tfl.cumsum"(%arg, %axis_p3) {exclusive = false, reverse = false} : (tensor<1x2x1x3xf32>, tensor<i32>) -> tensor<1x2x1x3xf32>
  func.return %res_m4, %res_m3, %res_m2, %res_m1, %res_00, %res_p1, %res_p2, %res_p3 : tensor<1x2x1x3xf32>, tensor<1x2x1x3xf32>, tensor<1x2x1x3xf32>, tensor<1x2x1x3xf32>, tensor<1x2x1x3xf32>, tensor<1x2x1x3xf32>, tensor<1x2x1x3xf32>, tensor<1x2x1x3xf32>

  // CHECK: %[[v1:.*]] = "tf.BatchMatMulV2"(%[[v0]], %arg1) <{adj_x = false, adj_y = false}> : (tensor<2x5x1x7xf32>, tensor<2x5x7x3xf32>) -> tensor<2x5x1x3xf32>
  // CHECK: %[[v2:.*]] = "tf.Reshape"(%[[v1]], %[[cst_1]]) : (tensor<2x5x1x3xf32>, tensor<3xi64>) -> tensor<2x5x3xf32>
  // CHECK: return %[[v2]] : tensor<2x5x3xf32>
}