%0 = "tf.BatchMatMulV2"(%arg3, %arg1) {adj_x = false, adj_y = false} : (tensor<1x3xf32>, tensor<3x4xf32>) -> tensor<1x4xf32>
    %1 = arith.constant dense<[[2.3, 3.4, 4.5, 5.5]]> : tensor<1x4xf32>
    %2 = "tf.Add"(%0, %1) : (tensor<1x4xf32>, tensor<1x4xf32>) -> tensor<1x4xf32>
    %3 = tensor.cast %2 : tensor<1x4xf32> to tensor<1x?xf32>
    func.return %3 : tensor<1x?xf32>
}

  %cst0 = "tfl.pseudo_const" () {value = dense<0.0> : tensor<1x4xf32>} : () -> tensor<1x4xf32> loc("Const")
  %cst1 = "tfl.pseudo_const" () {value = dense<0.0> : tensor<1x4xf32>} : () -> tensor<1x4xf32> loc("Const")
  // expected-error @+1 {{'tfl.lstm' op inputs don't match with the dimensions.}}

  %0:3 = "tf.Split"(%arg0, %arg1) : (tensor<i32>, tensor<1x4x3x3xf32>) -> (tensor<1x4x3xf32>, tensor<1x4x3xf32>, tensor<1x4x3xf32>)
  func.return %0#0 : tensor<1x4x3xf32>

  // CHECK-LABEL: split
  // CHECK: "tfl.split"(%arg0, %arg1) <{num_splits = 3 : i32}> : (tensor<i32>, tensor<1x4x3x3xf32>) -> (tensor<1x4x3xf32>, tensor<1x4x3xf32>, tensor<1x4x3xf32>)
}

  %0 = "tf.Cast"(%arg0) : (tensor<!tf_type.resource<tensor<8x40xf32>>>) -> tensor<*x!tf_type.resource>
  %1 = "tf.ReadVariableOp"(%0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<*x!tf_type.resource>) -> tensor<8x40xf32>
  func.return %1: tensor<8x40xf32>

  func.return %0 : tensor<5x2x5x10xf32>
}

// -----

func.func @testBatchMatMulV2DynamicOutputBatchDimension(%lhs: tensor<1x2x5x10xf32>, %rhs: tensor<1x10xf32>) -> tensor<?x2x5x10xf32> {
  %0 = "tf.BatchMatMulV2"(%lhs, %rhs) : (tensor<1x2x5x10xf32>, tensor<1x10xf32>) -> tensor<?x2x5x10xf32>
  func.return %0 : tensor<?x2x5x10xf32>
}

// -----