%cst = arith.constant dense<1> : tensor<1xi32>
  %0 = "tfl.exp"(%arg0) : (tensor<8x128xf32>) -> tensor<8x128xf32>
  %1 = "tfl.sum"(%0, %cst) {keep_dims = true} : (tensor<8x128xf32>, tensor<1xi32>) -> tensor<8x1xf32>
  %2 = "tfl.div"(%0, %1) {fused_activation_function = "NONE"} : (tensor<8x128xf32>, tensor<8x1xf32>) -> tensor<8x128xf32>
  func.return %2 : tensor<8x128xf32>

// CHECK-SAME:                      %[[VAL_0:.*]]: tensor<8x128xf32>,
// CHECK-SAME:                      %[[VAL_1:.*]]: tensor<f32>) -> tensor<11x131xf32> {
// CHECK:           %[[VAL_2:.*]] = arith.constant dense<{{\[\[}}1, 2], [0, 3]]> : tensor<2x2xi64>
// CHECK:           %[[VAL_3:.*]] = "tf.PadV2"(%[[VAL_0]], %[[VAL_2]], %[[VAL_1]]) : (tensor<8x128xf32>, tensor<2x2xi64>, tensor<f32>) -> tensor<11x131xf32>

// tf.XlaReduceScatter legalization
//===----------------------------------------------------------------------===//
// CHECK-LABEL: func @xla_reduce_scatter
func.func @xla_reduce_scatter(%arg0: tensor<128x128xf32>) -> tensor<64x128xf32> {
    %cst = "tf.Const"() {value = dense<0> : tensor<i32>} : () -> tensor<i32>
    %cst_0 = "tf.Const"() {value = dense<[[0, 4], [1, 5], [2, 6], [3, 7]]> : tensor<4x2xi32>} : () -> tensor<4x2xi32>