func.func @my_cast(%arg0: tensor<2x3xf32>) -> tensor<2x3xi32> {
  %0 = "tf.MyCast"(%arg0) {Tout=i32} : (tensor<2x3xf32>) -> tensor<2x3xi32>
  func.return %0 : tensor<2x3xi32>

// CHECK-NEXT: %[[CAST:.*]] = "tf.RiscCast"(%arg0) {Tout = i32} : (tensor<2x3xf32>) -> tensor<*xi32>
// CHECK-NEXT: %[[ES:.*]] = "tf.EnsureShape"(%[[CAST]]) <{shape = #tf_type.shape<2x3>}> : (tensor<*xi32>) -> tensor<2x3xi32>
// CHECK-NEXT: return %[[ES]] : tensor<2x3xi32>
}

    } : (tensor<?x100352xf32>, tensor<100352x10xf32>) -> tensor<?x10xf32>
    %3 = "tf.Identity"(%2) {device = ""} : (tensor<?x10xf32>) -> tensor<?x10xf32>
    return %3 : tensor<?x10xf32>
  }

  // CHECK-LABEL: func.func private @composite_dot_general_fn_1
  func.func private @composite_dot_general_fn_1(%arg0: tensor<?x100352xf32>, %arg1: tensor<100352x10xf32>) -> tensor<?x10xf32> {
    // CHECK-NOT: tf.CustomAggregator

  } : (tensor<2x7xi64>) -> (tensor<2x7xi32>)
  %1 = "tf.Identity"(%0) : (tensor<2x7xi32>) -> tensor<2x7xi32>
  func.return %1 : tensor<2x7xi32>
}

// CHECK-LABEL: func @func
// CHECK-SAME: %arg0: tensor<2x7xi64> {mhlo.sharding = "\08\03\1A\03\01\01\02\22\02\00\01B\01\00"
func.func @func(%arg0: tensor<2x7xi64>) -> (tensor<2x7xi32>) {

// CHECK: return %[[identity]] : tensor<1x3x2x2xf32>

func.func @cast_bf16_matmul_to_fp32(%arg0: tensor<1x10xf32>) -> (tensor<1x2xf32>) {
  %cst = "tf.Const"() {device = "", value = dense<1.000000e+01> : tensor<10x2xbf16>} : () -> tensor<10x2xbf16>
  %0 = "tf.Cast"(%arg0) {Truncate = false, device = ""} : (tensor<1x10xf32>) -> tensor<1x10xbf16>

func.func @transpose_2d_identity() -> tensor<2x2xi32> {
  %cst = arith.constant dense<[[0, 1], [2, 3]]> : tensor<2x2xi32>
  %cst_perm = arith.constant dense<[0, 1]> : tensor<2xi32>

  // CHECK: %[[CST:.*]] = arith.constant dense<{{\[\[}}0, 1], {{\[}}2, 3]]> : tensor<2x2xi32>
  // CHECK: return %[[CST]]
  %0 = "tfl.transpose"(%cst, %cst_perm) : (tensor<2x2xi32>, tensor<2xi32>) -> tensor<2x2xi32>
  func.return %0 : tensor<2x2xi32>
}

// CHECK:           %[[VAL_14:.*]] = "tf.Abs"(%[[VAL_1]]) : (tensor<2x3xi32>) -> tensor<2x3xi32>
// CHECK:           %[[VAL_15:.*]] = "tf.Div"(%[[VAL_13]], %[[VAL_14]]) : (tensor<2x3xi32>, tensor<2x3xi32>) -> tensor<2x3xi32>
// CHECK:           %[[VAL_16:.*]] = "tf.Select"(%[[VAL_6]], %[[VAL_7]], %[[VAL_15]]) : (tensor<2x3xi1>, tensor<2x3xi32>, tensor<2x3xi32>) -> tensor<2x3xi32>

// -----

// CHECK-LABEL: func @testBatchMatMulV2NoBatchDimension
func.func @testBatchMatMulV2NoBatchDimension(%lhs: tensor<5x10xf32>, %rhs: tensor<10x10xf32>) -> (tensor<5x10xf32>) {
  %0 = "tf.BatchMatMulV2"(%lhs, %rhs) : (tensor<5x10xf32>, tensor<10x10xf32>) -> tensor<5x10xf32>
  func.return %0 : tensor<5x10xf32>
}

// -----

// CHECK-LABEL: func @testBatchMatMulV2ValidBroadcastingBatchDimension

  %2 = "tfl.relu"(%arg0) : (tensor<1xf32>) -> tensor<1xf32>
  // CHECK: tac.device = "CPU", tac.inference_type = "FLOAT"
  %3 = "tfl.pack"(%arg0, %arg1) {axis = 0 : i32, values_count = 2 : i32} : (tensor<1xf32>, tensor<1xf32>) -> tensor<2x1xf32>
  func.return
}

func.func @notAnnotateConst(%arg0: tensor<256x32x32x3xf32>) -> tensor<256x30x30x16xf32> {
  // CHECK-NOT: tac.device tac.inference_type

  %1 = "tf.Concat"(%0, %arg0, %arg1) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<i32>, tensor<2x1xi32>, tensor<2x1xi32>) -> tensor<2x2xi32>
  func.return %1 : tensor<2x2xi32>
}

// CHECK-LABEL: testConcatCwiseUnary
func.func @testConcatCwiseUnary(%arg0: tensor<?x1xf32>, %arg1: tensor<?x1xf32>, %arg2: tensor<i32>) -> tensor<?x2xf32> {

    %1 = "tf.StatefulPartitionedCall"(%arg0) {config = "", config_proto = "", executor_type = "", f = @stateful_partitioned_call_func} : (tensor<20xi32>) -> tensor<*xi32>
    // CHECK: tf.TPUPartitionedCall
    // CHECK-SAME: (tensor<20xi32>, tensor<?xi32>) -> tensor<20xi32>