%11 = "mhlo.dot"(%10, %5) : (tensor<32x3920xf32>, tensor<3920x10xf32>) -> tensor<32x10xf32>
  %12 = "mhlo.broadcast_in_dim"(%2) <{broadcast_dimensions = dense<1> : tensor<1xi64>}> : (tensor<10xf32>) -> tensor<32x10xf32>
  %13 = mhlo.add %11, %12 : tensor<32x10xf32>
  %14 = mhlo.maximum %13, %0 : tensor<32x10xf32>
  return %14 : tensor<32x10xf32>
}
// CHECK: ReadVariableOp
// CHECK: mhlo.convolution
// CHECK: mhlo.add

func.func @testFullyConnectedWithBadInputShape(%arg0: tensor<2x2x11xf32>, %arg1: tensor<40x40xf32>, %arg2: none) -> tensor<40xf32> {
  // expected-error @+1 {{expect 'input' num_elements % 40 == 0, got input type 'tensor<2x2x11xf32>'}}
  %0 = "tfl.fully_connected"(%arg0, %arg1, %arg2) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<2x2x11xf32>, tensor<40x40xf32>, none) -> tensor<1x40xf32>

// CHECK-NEXT:    return %2 : tensor<3x2xf32>
}

func.func @transpose_3d(%arg0: tensor<1x2x3xf32>) -> tensor<3x2x1xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[2, 1, 0]> : tensor<3xi64>}> : (tensor<1x2x3xf32>) -> tensor<3x2x1xf32>
  func.return %0 : tensor<3x2x1xf32>

// CHECK-LABEL:   transpose_3d
// CHECK-NEXT:    %0 = "tfl.pseudo_const"() <{value = dense<[2, 1, 0]> : tensor<3xi64>}> : () -> tensor<3xi64>

    %2 = "quantfork.stats"(%1) {layerStats = dense<[5.00000000e-6, 7.00000000e-1]> : tensor<2xf32>} : (tensor<2x2x3xf32>) -> tensor<2x2x3xf32>
    return %2 : tensor<2x2x3xf32>
  }
// CHECK: func.func private @quantize_dot_general_batch_per_tensor_quantized_fn(%[[ARG_0:.+]]: tensor<2x2x2xf32>) -> tensor<2x2x3xf32> attributes {tf._original_func_name = "main_0"}

      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"

  %7 = "tf.BatchMatMulV2"(%1, %6) {adj_x = false, adj_y = false} : (tensor<8x8x40xf32>, tensor<40x10xf32>) -> tensor<8x8x10xf32>
  %8 = "tf.Add"(%7, %arg1) : (tensor<8x8x10xf32>, tensor<8x10xf32>) -> tensor<8x8x10xf32>
  %9 = "tf.Add"(%7, %arg2) : (tensor<8x8x10xf32>, tensor<8x10xf32>) -> tensor<8x8x10xf32>
  %10 = "tf.Add"(%arg1, %arg2) : (tensor<8x10xf32>, tensor<8x10xf32>) -> tensor<8x10xf32>

    : (tensor<?x3x2x1xf32>, tensor<2x1x1x1xf32>) -> tensor<?x2x2x1xf32>
  %1 = chlo.broadcast_add %0, %zp_offset : (
      tensor<?x2x2x1xf32>, tensor<?x2x2x1xf32>) -> tensor<?x2x2x1xf32>
  %2 = chlo.broadcast_add %1, %bias : (
      tensor<?x2x2x1xf32>, tensor<1xf32>) ->tensor<?x2x2x1xf32>
  return %2 : tensor<?x2x2x1xf32>
}

// -----

// CHECK-LABEL: func @dot_general_add_add

}

// -----

func.func @trivial_dynamic_update_slice(%arg0: tensor<2x7x14xf32>, %arg1: tensor<2x7x14xf32>) -> tensor<2x7x14xf32> {
  %0 = arith.constant dense<0> : tensor<3xi32>
  %1 = "tfl.dynamic_update_slice"(%arg0, %arg1, %0) : (tensor<2x7x14xf32>, tensor<2x7x14xf32>, tensor<3xi32>) -> tensor<2x7x14xf32>
  // CHECK: return %arg1
  func.return %1 : tensor<2x7x14xf32>
}

// -----

  // CHECK-LABEL: @get_next
  func.func private @get_next(%arg0: tensor<!tf_type.resource>) -> tensor<200x10xf32> {
    // CHECK: %0 = "tf.Iterator"
    // CHECK-SAME: shared_name = "foo_iterator"
    // CHECK: "tf.IteratorGetNext"(%0)
    %0 = "tf.IteratorGetNext"(%arg0) : (tensor<!tf_type.resource>) -> tensor<200x10xf32>
    return %0 : tensor<200x10xf32>
  }

  // CHECK-LABEL: @main
  func.func @main()

  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";