%5 = stablehlo.reshape %4 : (tensor<100xi64>) -> tensor<100x1xi64>
    %6 = "stablehlo.scatter"(%arg0, %5, %arg3) ({
    ^bb0(%arg5: tensor<f32>, %arg6: tensor<f32>):
      stablehlo.return %arg6 : tensor<f32>

  %5 = "tfl.reshape"(%4, %0) : (tensor<1x112x112x32x!quant.uniform<u8:f32, 0.023528476789885875>>, tensor<2xi32>) -> tensor<1x401408x!quant.uniform<u8:f32, 0.023528476789885875>>
  %6 = "tfl.softmax"(%5) {beta = 1.000000e+00 : f32} : (tensor<1x401408x!quant.uniform<u8:f32, 0.023528476789885875>>) -> tensor<1x401408x!quant.uniform<u8:f32, 3.906250e-03>>

  %1 = "tf.ReadVariableOp"(%0) {device = "/device:CPU:0"} : (tensor<!tf_type.resource<tensor<f32>>>) -> tensor<f32>
  %2 = "tf.XlaLaunch"(%arg0, %1) {_noinline = true, _xla_compile_device_type = "GPU", device = "/device:GPU:0", function = @xla_func_0, operandSegmentSizes = array<i32: 0, 2, 0>} : (tensor<7xf32>, tensor<f32>) -> tensor<10xf32>
  func.return %2 : tensor<10xf32>

    "tfl.yield"(%barg0, %cst) : (tensor<f32>, tensor<f32>) -> ()
  }) : (tensor<f32>, tensor<f32>) -> (tensor<f32>, tensor<f32>)
  func.return %0#1 : tensor<f32>
}

// -----

// Test case to test While op with resources that are not read-only variables.
// Do not remove resource arugments if they are not read-only variables to keep
// the graph's control dependency.
// CHECK-LABEL: WhileWithNonReadOnlyVariableResources

// CHECK-DAG:       %[[VAL_1:.*]] = "tf.Const"() <{value = dense<1.230000e+02> : tensor<f32>}> : () -> tensor<f32>
// CHECK-DAG:       %[[VAL_2:.*]] = "tf.Const"() <{value = dense<1.000000e+00> : tensor<f32>}> : () -> tensor<f32>
// CHECK:           %[[VAL_3:.*]] = "tf.Range"(%[[VAL_0]], %[[VAL_1]], %[[VAL_2]]) : (tensor<f32>, tensor<f32>, tensor<f32>) -> tensor<123xf32>

// CHECK-SAME: ([[ARG0:%.*]]: tensor<i32>, [[ARG1:%.*]]: tensor<f32>)
func.func @testIdentityN(%arg0: tensor<i32>, %arg1: tensor<f32>) {
  // CHECK-NOT:  "tf.IdentityN"
  // CHECK:      "tf_device.cluster"
  // CHECK-NEXT: tf_device.return [[ARG0]], [[ARG1]]
  %0:2 = "tf_device.cluster"() ({
    %1:2 = "tf.IdentityN"(%arg0, %arg1) : (tensor<i32>, tensor<f32>) -> (tensor<i32>, tensor<f32>)
    tf_device.return %1#0, %1#1 : tensor<i32>, tensor<f32>

  %shape = arith.constant dense<[6, 4]> : tensor<2xi32>
  %0 = "tfl.reshape"(%arg0, %shape) : (tensor<1x2x3x4x!quant.uniform<i8:f32, 3.0>>, tensor<2xi32>) -> tensor<6x4x!quant.uniform<i8:f32, 3.0>>

// RUN: tf-opt %s --tf-print | FileCheck %s

module {
// Smoke test. We don't expect any modifications of the MLIR.

// CHECK-LABEL: foo
// CHECK: return
func.func @foo(%arg0: tensor<f32>, %arg1: tensor<f32>) -> tensor<f32> {
  return %arg0 : tensor<f32>
}

    func.return %0 : tensor<?x19xf32>
  }
}

// CHECK-LABEL: HloModule main
// CHECK:       (arg_tuple.{{[0-9]+}}: (f32[10,17], f32[17,19])) -> (f32[10,19])

// NO_TUPLES-LABEL: HloModule main

func.func @mul_complex32(%arg0: tensor<14xcomplex<f32>>, %arg1: tensor<14xcomplex<f32>>) -> tensor<14xcomplex<f32>> {
  %0 = "tf.Mul"(%arg0, %arg1) : (tensor<14xcomplex<f32>>, tensor<14xcomplex<f32>>) -> tensor<14xcomplex<f32>>
  func.return %0: tensor<14xcomplex<f32>>

// CHECK-LABEL: mul_complex32
// CHECK:  tfl.mul %arg0, %arg1 {fused_activation_function = "NONE"} : tensor<14xcomplex<f32>>
// CHECK:  return
}