"tf_device.cluster_func"(%1) {func = @cluster_func, use_spmd_for_xla_partitioning = true, num_cores_per_replica = 8 : i64} : (tensor<4x4x4xf32>) -> ()
  func.return
}

func.func @cluster_func(%arg0: tensor<4x4x4xf32>) {
  func.return
}

// -----

// Tests partitioned inputs/outputs with no sharding (via XLA SPMD) defaults to
// replicate sharding ("").

// CHECK-LABEL: func @partitioned_input_output

  module {
    func.func @main(%arg0: tensor<2x2x2xf32>, %arg1: tensor<2x2x2xf32>) -> tensor<2x2x2xf32> attributes {_from_xla_call_module} {
      %0 = stablehlo.dot_general %arg0, %arg1,
        batching_dims = [0] x [0],
        contracting_dims = [2] x [1],
        precision = [DEFAULT, DEFAULT]
      : (tensor<2x2x2xf32>, tensor<2x2x2xf32>) -> tensor<2x2x2xf32>
        return %0 : tensor<2x2x2xf32>
    }
  }
)mlir";

  %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>

  constexpr absl::string_view kModuleGather = R"mlir(
    module {
      func.func @main() -> (tensor<2x3x2x2xf32>) {
        %0 = stablehlo.constant dense<1.0> : tensor<3x4x2xf32>
        %1 = stablehlo.constant dense<2> : tensor<2x3x2xi64>
        %2 = "stablehlo.gather"(%0, %1) {
          dimension_numbers = #stablehlo.gather<
            offset_dims = [2, 3],

func.func @leaky_relu_grad(%arg0: tensor<1x4x4xf32>, %arg1: tensor<1x4x4xf32>) -> tensor<1x4x4xf32> attributes {tf.entry_function = {}} {
    // CHECK-NEXT: %[[ALPHA:.*]] = "chlo.constant_like"(%arg1) <{value = 2.000000e-01 : f32}> : (tensor<1x4x4xf32>) -> tensor<1x4x4xf32>
    // CHECK-NEXT: %[[ZERO:.*]] = "chlo.constant_like"(%arg1) <{value = 0.000000e+00 : f32}> : (tensor<1x4x4xf32>) -> tensor<1x4x4xf32>

func.func @no_qdq_except_conv_and_dot_general(%arg0: tensor<2x3x2xi64>) -> tensor<2x3x2x2xf32> attributes {tf._original_func_name = "main_0"} {
  %cst = "tf.Const"() {value = dense<3.000000e-01> : tensor<3x4x2xf32>} : () -> tensor<3x4x2xf32>
  %0 = "tf.XlaCallModule"(%cst, %arg0) {
    Sout = [#tf_type.shape<1x3>], _entry_function = @composite_gather_fn,
    _original_entry_function = "composite_gather_fn", _quantization_method = "weight_only_ptq { }",

  // CHECK: %[[v1:.*]] = "tf.BatchMatMulV2"(%arg0, %[[v0]]) <{adj_x = false, adj_y = false}> : (tensor<5x4x3xf32>, tensor<1x3x2xf32>) -> tensor<5x4x2xf32>
  // CHECK: %[[v2:.*]] = "tf.Transpose"(%[[v1]], %[[cst_1]]) : (tensor<5x4x2xf32>, tensor<3xi32>) -> tensor<4x2x5xf32>
  // CHECK: return %[[v2]] : tensor<4x2x5xf32>

  func.return %0 : tensor<4x4x4xf32>

  // CHECK-DAG: %[[CST:.*]] = arith.constant dense<[0, 1, 0]> : tensor<3xi32>
  // CHECK-DAG: %[[CST_0:.*]] = arith.constant dense<0> : tensor<3xi32>
  // CHECK-DAG: %[[CST_1:.*]] = arith.constant dense<1> : tensor<3xi32>

  // expected-error @+1 {{op requires indices operand to have at least 1 dimension}}
  %0 = "tf.TensorScatterUpdate"(%tensor, %indices, %updates) : (tensor<4x4x4xf32>, tensor<i32>, tensor<4x4xf32>) -> tensor<4x4x4xf32>
  func.return %0 : tensor<4x4x4xf32>
}

// -----

func.func @tensor_scatter_update(%tensor: tensor<4x4x4xf32>, %indices: tensor<4x2xi32>, %updates: tensor<f32>) -> tensor<4x4x4xf32> {

}

// -----

// CHECK-LABEL: topk_2
func.func @topk_2(%arg0: tensor<3x4x8xf32>) -> (tensor<3x4x2xf32>, tensor<3x4x2xi32>) {
  %0 = arith.constant dense<2> : tensor<i32>
  %1:2 = "tfl.topk_v2"(%arg0, %0) : (tensor<3x4x8xf32>, tensor<i32>) -> (tensor<3x4x2xf32>, tensor<3x4x2xi32>)
  func.return %1#0, %1#1: tensor<3x4x2xf32>, tensor<3x4x2xi32>
}

// -----

// CHECK-LABEL: topk_d