// CHECK-NEXT:  %[[cst:.*]] = arith.constant dense<1> : tensor<i32>
  %cst = arith.constant dense<1> : tensor<i32>
// CHECK-NEXT:  %[[softmax:.*]] = "tfl.softmax"(%arg0) <{beta = 1.000000e+00 : f32}> : (tensor<128x16xf32>) -> tensor<128x16xf32>
  %0 = "tfl.softmax"(%arg0) {beta = 1.000000e+00 : f32} : (tensor<128x16xf32>) -> tensor<128x16xf32>

}

// -----

module {
func.func private @func_20_GPU_FLOAT(%arg0: tensor<128x128xf32>, %arg1: tensor<3xi32>) -> tensor<1x128x128xf32> attributes {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_20"} {
  %0 = "tfl.reshape"(%arg0, %arg1) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<128x128xf32>, tensor<3xi32>) -> tensor<1x128x128xf32>
  func.return %0 : tensor<1x128x128xf32>
}

   "tfl.yield"(%4) : (tensor<128x128xf32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>

   "tfl.yield"(%4) : (tensor<128x128xf32>) -> ()
  }) {_tfl_quant_trait = "fully_quantizable", device = ""} : (tensor<128x128xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xi32>) -> tensor<128x128xf32>

func.func @softmax(%arg0: tensor<8x16xf32>) -> tensor<8x16xf32> {
  %0 = "tf.Softmax"(%arg0) : (tensor<8x16xf32>) -> tensor<8x16xf32>
  func.return %0 : tensor<8x16xf32>
// CHECK: %[[CUSTOM_0:.*]] = "tfl.custom"(%arg0) <{custom_code = "FlexSoftmax", custom_option = #tfl<const_bytes : "0x07536F66746D617800161207536F66746D61781A002A070A0154120230013200000221191414042801">}> : (tensor<8x16xf32>) -> tensor<8x16xf32>

  %0 = "tf.GatherV2"(%cst_0, %arg0, %cst) {batch_dims = 0 : i64, device = ""} : (tensor<128x32xf32>, tensor<6xi64>, tensor<i32>) -> tensor<6x32xf32>
  return %0 : tensor<6x32xf32>

// CHECK-DAG: %[[CST:.*]] = "tf.Const"() {{.*}} : () -> tensor<i32>
// CHECK-DAG: %[[CST_1:.*]] = "tf.Const"() {{.*}} : () -> tensor<128x32xf32>

// CHECK: return %[[v4]] : tensor<?x?x8x?xf32>
}

func.func @einsum_ellipsis(%arg0: tensor<1x512x128xf32>, %arg1: tensor<128x256xf32>) -> tensor<1x512x256xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {device = "", equation = "...x,xy->...y"} : (tensor<1x512x128xf32>, tensor<128x256xf32>) -> tensor<1x512x256xf32>
  func.return %0 : tensor<1x512x256xf32>
// CHECK-LABEL: einsum_ellipsis

  // 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>

  %max_size = "tf.Const"() {value = dense<10> : tensor<i32>} : () -> tensor<i32>
  // CHECK-NOT: tf.EmptyTensorList
  // CHECK: tf.BroadcastTo
  // CHECK-SAME: tensor<10x16xf32>
  %tl0 = "tf.EmptyTensorList"(%element_shape, %max_size) : (tensor<1xi32>, tensor<i32>) -> tensor<!tf_type.variant<tensor<16xf32>>>
  // CHECK-NOT: tf.TensorListReserve
  // CHECK: tf.BroadcastTo