%5 = "quantfork.qcast"(%4) {volatile} : (tensor<3x4x2xf32>) -> tensor<3x4x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>
    %6 = "quantfork.dcast"(%5) : (tensor<3x4x2x!quant.uniform<i8:f32, 0.13170163023705575:-1>>) -> tensor<3x4x2xf32>
    %7 = "stablehlo.gather"(%6, %arg2) {
    dimension_numbers = #stablehlo.gather<

// CHECK-NEXT: return %[[dq]] : tensor<2x2xf32>
}

// CHECK-LABEL: prepareStatistics
func.func @prepareStatistics(%arg0: tensor<8x4x3xf32>) -> tensor<8x4x3xf32> {
  %0 = "quantfork.stats"(%arg0) {
    layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>
  } : (tensor<8x4x3xf32>) -> tensor<8x4x3xf32>
  %1 = "quantfork.stats"(%0) {
    layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>,
    axisStats = dense<[
      [-1.0, 1.0],

//CHECK:func.func private @slice(%arg0: tensor<160x20x1xf32>) -> tensor<1x1x1xf32> {
//CHECK-NEXT: %0 = "vhlo.slice_v1"(%arg0) <{limit_indices = #vhlo.tensor_v1<dense<0> : tensor<3xi64>>, start_indices = #vhlo.tensor_v1<dense<1> : tensor<3xi64>>, strides = #vhlo.tensor_v1<dense<1> : tensor<3xi64>>}> : (tensor<160x20x1xf32>) -> tensor<1x1x1xf32> 
//CHECK-NEXT: return %0 : tensor<1x1x1xf32>
//CHECK-NEXT:}

  %0 = "tfl.var_handle"() {container = "", shared_name = "states"} : () -> tensor<!tf_type.resource<tensor<1x2x3xf32>>>
  %1 = "tfl.read_variable"(%0) : (tensor<!tf_type.resource<tensor<1x2x3xf32>>>) -> tensor<1x2x3xf32>
  %2 = "tfl.quantize"(%1) {qtype = tensor<1x2x3x!quant.uniform<u8<1:255>:f32, 0.047244094488188976:128>>} : (tensor<1x2x3xf32>) -> tensor<1x2x3x!quant.uniform<u8<1:255>:f32, 0.047244094488188976:128>>

func.func @lower_einsum(%arg0: tensor<3x4x5xf32>, %arg1: tensor<3x5x6xf32>) -> tensor<3x4x6xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "ijk,ikm->ijm"}: (tensor<3x4x5xf32>, tensor<3x5x6xf32>) -> tensor<3x4x6xf32>
  func.return %0 : tensor<3x4x6xf32>
}
// CHECK-LABEL: lower_einsum
// CHECK: "tf.BatchMatMulV2"(%arg0, %arg1) <{adj_x = false, adj_y = false}> : (tensor<3x4x5xf32>, tensor<3x5x6xf32>) -> tensor<3x4x6xf32>

  func.func @batchmatmulv2(%arg0: tensor<1x4x2xf32>, %arg1: tensor<3x2x4xf32>) -> tensor<3x4x4xf32> {
    // CHECK: mhlo.reduce
    // CHECK: mhlo.dot_general
    // CHECK: mhlo.transpose
    %0 = "tf.BatchMatMulV2"(%arg0, %arg1) {T = f32, adj_x = false, adj_y = false, grad_x = false, grad_y = false, device = ""} : (tensor<1x4x2xf32>, tensor<3x2x4xf32>) -> tensor<3x4x4xf32>
    func.return %0 : tensor<3x4x4xf32>
  }

func.func @partitioned_input_rank_mismatch(%arg0: tensor<!tf_type.resource<tensor<1x4x4xf32>>>) {
  // expected-error @+1 {{rank}}
  %0 = "tf.TPUPartitionedInputV2"(%arg0) {_XlaSharding = "\08\03\1A\05\04\01\01\01\02\22\08\00\01\02\03\04\05\06\070\01", partition_dims = [4, 1, 1, 2], is_packed = true} : (tensor<!tf_type.resource<tensor<1x4x4xf32>>>) -> tensor<!tf_type.resource<tensor<4x4x4xf32>>>