%axis_p3 = arith.constant dense<3> : tensor<i32>
  %res_m4 = "tfl.cumsum"(%arg, %axis_m4) {exclusive = false, reverse = false} : (tensor<1x2x1x3xf32>, tensor<i32>) -> tensor<1x2x1x3xf32>  // Eliminated
  %res_m3 = "tfl.cumsum"(%arg, %axis_m3) {exclusive = false, reverse = false} : (tensor<1x2x1x3xf32>, tensor<i32>) -> tensor<1x2x1x3xf32>

  // CHECK: mhlo.constant dense<> : tensor<0xf32>
  %1 = "tf.Cumsum"(%arg0, %0) : (tensor<0xf32>, tensor<i32>) -> tensor<0xf32>
  func.return %1 : tensor<0xf32>
}

// -----

// CHECK-LABEL: func @cumsum_dynamic
func.func @cumsum_dynamic(%arg0: tensor<?xf32>, %arg1: tensor<i32>) -> tensor<?xf32> {
  // CHECK: "tf.Cumsum"
  %0 = "tf.Cumsum"(%arg0, %arg1) : (tensor<?xf32>, tensor<i32>) -> tensor<?xf32>

  %213 = "tf.Mul"(%212, %12) {device = ""} : (tensor<i64>, tensor<1xi64>) -> tensor<1xi64>
  %214 = "tf.Tile"(%213, %211) {device = ""} : (tensor<1xi64>, tensor<1xi64>) -> tensor<?xi64>
  %215 = "tf.Cumsum"(%214, %14) {device = "", exclusive = false, reverse = false} : (tensor<?xi64>, tensor<i32>) -> tensor<?xi64>
  %216 = "tf.ConcatV2"(%11, %215, %3) {device = ""} : (tensor<1xi64>, tensor<?xi64>, tensor<i32>) -> tensor<?xi64>

// CHECK-DAG:       %[[VAL_2:.*]] = "tf.Const"() <{value = dense<0> : tensor<i64>}> : () -> tensor<i64>
// CHECK:           %[[VAL_3:.*]] = "tf.Cumsum"(%[[VAL_0]], %[[VAL_2]]) <{exclusive = false, reverse = false}> : (tensor<4x12xf32>, tensor<i64>) -> tensor<4x12xf32>
// CHECK:           return %[[VAL_3]] : tensor<4x12xf32>
// CHECK:         }

    if (result.getType() != op.getType())
      result = rewriter.create<tensor::CastOp>(loc, op.getType(), result);
    rewriter.replaceOp(op, result);
    return success();
  }
};

// Converts the Cumsum or Cumprod TensorFlow op to the HLO ReduceWindow op by
// setting appropriate window dimensions, with the given aggregation op as the
// reduction function. The input tensor needs to have a static shape, and 'axis'