%2 = "tf.Const"() {value = dense<0.0> : tensor<2x2xf32>} : () -> tensor<2x2xf32>
  %3 = "tf.Const"() {value = dense<0.0> : tensor<4x4xf32>} : () -> tensor<4x4xf32>
  "tf.SendTPUEmbeddingGradients"(%2, %3) {_tpu_embedding_layer = "call1", config = "\0A\0B\0C\0D", operandSegmentSizes = array<i32: 2, 0>} : (tensor<2x2xf32>, tensor<4x4xf32>) -> ()

  func.return %2 : tensor<4x4x4xf32>
}

  %0 = "tfl.pseudo_const" () {value = dense<0.0> : tensor<4x4xf32>} : () -> tensor<4x4xf32> loc("Const")
  %1 = "tfl.unidirectional_sequence_rnn"(%arg0, %arg1, %arg2, %arg3, %0) {fused_activation_function = "TANH", time_major = true} : (tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4x4xf32>) -> tensor<4xf32>
  func.return %1 : tensor<4xf32>

func.func @test_batch_matmul_to_einsum(%arg0: tensor<1x2x3xf32>, %arg1: tensor<3x4xf32>) -> tensor<1x2x4xf32> {
  // CHECK-LABEL: test_batch_matmul_to_einsum
  // CHECK: "tf.Einsum"(%arg0, %arg1) <{equation = "...mk,...kn->...mn"}> : (tensor<1x2x3xf32>, tensor<3x4xf32>) -> tensor<1x2x4xf32>
  %0 = "tf.BatchMatMul"(%arg0, %arg1) {adj_x = false, adj_y = false} : (tensor<1x2x3xf32>, tensor<3x4xf32>) -> tensor<1x2x4xf32>
  func.return %0: tensor<1x2x4xf32>
}

func.func @foo(%arg0: tensor<1x4xf32>, %arg1: tensor<1x4xf32>) -> tensor<1x4xf32> {
  func.return %arg0 : tensor<1x4xf32>
}

func.func @bar(%arg0: tensor<2x4xf32>, %arg1: tensor<2x4xf32>) -> tensor<2x4xf32> {
  func.return %arg0 : tensor<2x4xf32>
}

func.func @foobar(%arg0: tensor<1x4xf32>, %arg1: tensor<1x4xf32>) -> tensor<1x4xf32> {
  func.return %arg0 : tensor<1x4xf32>
}

// CHECK-DAG: func @main

func.func @main(%arg0: tensor<1x4xf32>) -> tensor<1x3xf32> {
  %0 = stablehlo.constant dense<1.0> : tensor<4x3xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
  return %1 : tensor<1x3xf32>
}
// CHECK: @main(%[[ARG_0:.+]]: tensor<1x4xf32>) -> tensor<1x3xf32>

// CHECK: return

  func.func private @composite_dot_general_fn(%arg0: tensor<1x4xf32>, %arg1: tensor<4x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {
      %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
      return %0 : tensor<1x3xf32>
  }
}

// -----

  // CHECK-SAME: %[[ARG3:[^:[:space:]]+]]
  func.func private @main(%arg0: tensor<1x4xf32>, %arg1: tensor<4x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {
    %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
    return %0 : tensor<1x3xf32>
    // CHECK: %[[DOT:.+]] = stablehlo.dot_general %[[ARG2]], %[[ARG3]]
    // CHECK: return %[[DOT]]
  }

  %0 = "tf.BatchMatMulV2"(%arg0, %arg1) {T = f32, adj_x = false, adj_y = false, device = ""} : (tensor<3x4x2xf32>, tensor<2x4xf32>) -> tensor<3x4x4xf32>
  func.return %0 : tensor<3x4x4xf32>
}

func.func @batchmatmulv2_rhs_batch(%arg0: tensor<4x2xf32>, %arg1: tensor<3x2x4xf32>) -> tensor<3x4x4xf32> {
// CHECK-LABEL:   func @batchmatmulv2_rhs_batch

    module attributes {} {
      func.func @main(%arg0: tensor<1x4xf32>) -> tensor<1x3xf32> attributes {} {
        %0 = stablehlo.constant dense<1.0> : tensor<4x3xf32>
        %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
        return %1 : tensor<1x3xf32>
      }
    }
  )mlir");
  ASSERT_TRUE(module_op);