// CHECK-DAG: %[[CST_2:.*]] = "tf.Const"() <{value = dense<1.38408304E-5> : tensor<f32>}> : () -> tensor<f32>
// CHECK-DAG: %[[CST_3:.*]] = "tf.Const"() <{value = dense<0> : tensor<i32>}> : () -> tensor<i32>
// CHECK-DAG: %[[CST_4:.*]] = "tf.Const"() <{value = dense<0.0027450982> : tensor<f32>}> : () -> tensor<f32>

  %0:4 = "tf.CustomAggregator"(%arg0) {min = -0.1 : f32, max = 0.2 : f32, id = "0", calibration_method = 1 : i32, num_bins = 0 : i32, max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32} : (tensor<8x8x8x8xf32>) -> (tensor<8x8x8x8xf32>, tensor<f32>, tensor<f32>, tensor<*xi64>)

  // CHECK-NEXT:   %[[ID:.*]] = "tf_device.launch"() <{device = "TPU_REPLICATED_HOST_0"}> ({
  // CHECK-NEXT:     %[[IDINSIDE:.*]] = "tf.Identity"(%[[REPVAR]]) {ici_weight_distribution_mlir_bridge_marker = true} : (tensor<f32>) -> tensor<f32>
  // CHECK-NEXT:     tf_device.return %[[IDINSIDE]] : tensor<f32>
  // CHECK-NEXT:   }) : () -> tensor<f32>
  // CHECK-NEXT:   "tf_device.cluster"() ({

}

func.func @batchmatmulv2_adj_complex(%arg0: tensor<2x5xcomplex<f32>>, %arg1: tensor<4x2xcomplex<f32>>) -> tensor<5x4xcomplex<f32>> {
// CHECK-LABEL:   func @batchmatmulv2_adj_complex(
// CHECK-SAME:                                    [[LHS:%.*]]: tensor<2x5xcomplex<f32>>, [[RHS:%.*]]: tensor<4x2xcomplex<f32>>) -> tensor<5x4xcomplex<f32>> {
// CHECK:           [[LHSRE:%.*]] = mhlo.real [[LHS]]

func.func @allowsMHLO() -> (tensor<8x64x32x4xcomplex<f32>> {mhlo.sharding = ""}) {
  %0 = mhlo.constant dense<(1.000000e+00,-1.000000e+00)> : tensor<128x32x4xcomplex<f32>>
  %1 = mhlo.constant dense<(1.000000e+00,1.000000e+00)> : tensor<8x64x128xcomplex<f32>>
  %2 = "mhlo.einsum"(%1, %0) <{einsum_config = "abc,cde->abde"}> : (tensor<8x64x128xcomplex<f32>>, tensor<128x32x4xcomplex<f32>>) -> tensor<8x64x32x4xcomplex<f32>>
  return %2 : tensor<8x64x32x4xcomplex<f32>>
}

  %input_zps = "tf.Const"() { value = dense<4> : tensor<i32> } : () -> tensor<i32>
  %bias_scales = "tf.Const"() { value = dense<2.0> : tensor<f32> } : () -> tensor<f32>
  %bias_zps = "tf.Const"() { value = dense<4> : tensor<i32> } : () -> tensor<i32>
  %output_scales = "tf.Const"() { value = dense<2.0> : tensor<f32> } : () -> tensor<f32>

  %min = arith.constant dense<0.0> : tensor<f32>
  %max = arith.constant dense<255.0> : tensor<f32>
  %mini = "tf.Identity"(%min) : (tensor<f32>) -> tensor<f32>
  %maxi = "tf.Identity"(%max) : (tensor<f32>) -> tensor<f32>
  %rst = "tf.FakeQuantWithMinMaxVars"(%in, %mini, %maxi) {num_bits = 5, narrow_range = false} : (tensor<8xf32>, tensor<f32>, tensor<f32>) -> tensor<8xf32>
  func.return %rst : tensor<8xf32>

   %0 = "tfl.pseudo_const"() {value = dense<[1, 2]> : tensor<2xi32>} : () -> tensor<2xi32>
   %1 = "tfl.mean"(%arg0, %0) {keep_dims = false} : (tensor<?x7x7x2048x!quant.uniform<i8:f32, 0.6:-128>>, tensor<2xi32>) -> tensor<?x2048x!quant.uniform<i8:f32, 0.6:-128>>
  func.return %1 : tensor<?x2048x!quant.uniform<i8:f32, 0.6:-128>>
}

// CHECK: %[[id2:.*]] = "tf.RiscSame"(%arg2) : (tensor<f32>) -> tensor<*xf32>
// CHECK: %[[es:.*]] = "tf.EnsureShape"(%[[id2]]) <{shape = #tf_type.shape<>}> : (tensor<*xf32>) -> tensor<f32>
// CHECK: return %[[es]] : tensor<f32>
}

// CHECK-LABEL: decompose_tf_concat_n
func.func @decompose_tf_concat_n(%arg0: tensor<f32>, %arg1: tensor<f32>, %arg2: tensor<f32>) -> tensor<3xf32> {
  %0 = "tfr.cast"(%arg0) : (tensor<f32>) -> !tfr.tensor

  %1 = stablehlo.constant dense<0x7FC00000> : tensor<f32>
  %2 = stablehlo.constant dense<6.000000e+00> : tensor<f32>
  %3 = stablehlo.constant dense<0.000000e+00> : tensor<f32>
  %4 = "stablehlo.add"(%0, %1) : (tensor<f32>, tensor<f32>) -> tensor<f32>
  %5 = stablehlo.clamp %3, %arg0, %2 : (tensor<f32>, tensor<?x112x112x64xf32>, tensor<f32>) -> tensor<?x112x112x64xf32>
  %6 = "stablehlo.reduce_window"(%5, %4) ({