// 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:  %5 = "tfl.arg_max"(%arg0, %cst) : (tensor<1x32x1xf32>, tensor<1xi32>) -> tensor<1x1xi32>
  // CHECK:  return %4, %5 : tensor<1x1xf32>, tensor<1x1xi32>
}

// CHECK-LABEL:   func @convert_pytorch_argmax
func.func @convert_pytorch_argmax(%arg0: tensor<1x9xi32>) -> tensor<1xi32> {
  %0 = mhlo.constant dense<0> : tensor<i32>

  %3 = "tf.PreventGradient"(%1) : (tensor<10x10xi32>) -> tensor<10x10xi32>
  %4 = "tf.AddV2"(%2, %3) {device = ""} : (tensor<10x10xi32>, tensor<10x10xi32>) -> tensor<10x10xi32>
  func.return %4 : tensor<10x10xi32>

// CHECK-LABEL: preventGradient
// CHECK-NOT: %2 = "tf.PreventGradient"(%0) : (tensor<10x10xi32>) -> tensor<10x10xi32>

// CHECK:           %[[SUB:.*]] = "tf.Sub"(%[[CST]], %[[SCATTER]]) : (tensor<i32>, tensor<1x24xi32>) -> tensor<1x24xi32>
// CHECK:           %[[MUL:.*]] = "tf.Mul"(%[[SUB]], %[[CAST0]]) : (tensor<1x24xi32>, tensor<1x24xi32>) -> tensor<1x24xi32>

    %10 = stablehlo.dot_general %8, %9, contracting_dims = [1] x [0] : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
    %11 = stablehlo.convert %3 : (tensor<1x3xi32>) -> tensor<1x3xf32>
    %12 = stablehlo.subtract %10, %11 : tensor<1x3xf32>  // q1 * q2 - z1 * q2
    %13 = stablehlo.multiply %12, %4 : tensor<1x3xf32>  // s1 * s2

// CHECK-LABEL: fuseTileWithBinaryOp
func.func @fuseTileWithBinaryOp(%arg0: tensor<1x1xf32>) -> tensor<1x2xf32> {
  %cst = arith.constant dense<[[1,2]]> : tensor<1x2xi32>
  %cst1 = arith.constant dense<[[3.0, 4.0]]> : tensor<1x2xf32>
  %0 = "tfl.sqrt"(%arg0) : (tensor<1x1xf32>) -> tensor<1x1xf32>
  %1 = "tfl.tile"(%0, %cst) : (tensor<1x1xf32>, tensor<1x2xi32>) -> tensor<1x2xf32>

// NOFALLBACK-LABEL: @xla_svd
func.func @xla_svd(%arg0: tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>) {
  // NOFALLBACK: XlaSvd
  %s, %u, %v = "tf.XlaSvd"(%arg0) {max_iter = 1, epsilon = 1.0E-09 : f32, precision_config = ""} : (tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>)
  func.return %s, %u, %v : tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>
}

  %group_assignment = "tf.Const"() { value = dense<[[0],[1]]> : tensor<2x1xi32> } : () -> tensor<2x1xi32>
  // CHECK: "mhlo.all_reduce"
  // CHECK{LITERAL}: replica_groups = dense<[[0], [1]]> : tensor<2x1xi64>
  // CHECK-NOT: channel_handle
  // CHECK: mhlo.add
  %0 = "tf.XlaAllReduce"(%input, %group_assignment) {reduce_op = "Add", mode = "CrossReplica"} : (tensor<f32>, tensor<2x1xi32>) -> tensor<f32>
  func.return %0 : tensor<f32>
}

// -----

  %20 = "tf.ExpandDims"(%19, %7) {device = ""} : (tensor<1xi32>, tensor<i32>) -> tensor<1x1xi32>
  %21 = "tf.Cast"(%20) {Truncate = false, device = ""} : (tensor<1x1xi32>) -> tensor<1x1xi64>
  %22 = "tf.Reshape"(%21, %12) {device = ""} : (tensor<1x1xi64>, tensor<1xi64>) -> tensor<1xi64>
  %23 = "tf.Reshape"(%arg0, %5) {device = ""} : (tensor<1x!tf_type.string>, tensor<1xi32>) -> tensor<1x!tf_type.string>

}

func.func @addN(%arg0: tensor<2x3xi32>, %arg1: tensor<2x3xi32>, %arg2: tensor<2x3xi32>) -> tensor<2x3xi32> {
  %0 = "tf.AddN"(%arg0, %arg1, %arg2) : (tensor<2x3xi32>, tensor<2x3xi32>, tensor<2x3xi32>) -> tensor<2x3xi32>
  func.return %0 : tensor<2x3xi32>

// CHECK-LABEL: addN
// CHECK:  "tfl.add_n"(%arg0, %arg1, %arg2) : (tensor<2x3xi32>, tensor<2x3xi32>, tensor<2x3xi32>) -> tensor<2x3xi32>
// CHECK:  return
}