# CHECK-DAG: tf.IteratorGetNextSync{{.+}}-> (tensor<*xi16>, tensor<3x24xi32>, tensor<?x4x32xi64>)
# CHECK-DAG: tf.MultiDeviceIteratorGetNextFromShard{{.+}}-> (tensor<5x40xf16>, tensor<*xcomplex<f32>>, tensor<6x48x?xcomplex<f64>>)
# CHECK-DAG: tf.InfeedDequeueTuple{{.+}}-> (tensor<?x?x?xui16>, tensor<*xui32>, tensor<7x56xui64>)
# CHECK-DAG: tf.InfeedDequeue{{.+}}-> tensor<?x8x?xi8>

)

var (
	i8 int8 = 0x12
	i16 int16 = 0x1234
	i32 int32 = 0x12345678
	i64 int64 = 0x123456789abcdef0
	ui8 uint8 = 0x12
	ui16 uint16 = 0x1234
	ui32 uint32 = 0x12345678
	ui64 uint64 = 0x123456789abcdef0

	ni8 = ^i8
	ni16 = ^i16
	ni32 = ^i32
	ni64 = ^i64
	nui8 = ^ui8
	nui16 = ^ui16
	nui32 = ^ui32
	nui64 = ^ui64
)

var nfail = 0

func main() {
	if nfail > 0 {
		fmt.Printf("BUG\n")
	}
}

  %0 = "tf.Const"() {value = dense<[1, 2, 3, 4]> : tensor<4xui64>} : () -> tensor<4xui64>
  // CHECK: tfrt_fallback_async.const_dense_tensor  dense<1.000000e+00> : tensor<1xbf16>
  %1 = "tf.Const"() {device = "/device:CPU:0", value = dense<[1.0]> : tensor<1xbf16>} : () -> tensor<4xbf16>
  // CHECK: corert.executeop({{.*}}) "tf.Const"() {dtype = ui64, value = dense<[1, 2, 3, 4]> : tensor<4xui64>} : 1

func.func @preserve_assert(%arg0: tensor<1xi32>, %arg1: tensor<1xi32>) -> tensor<1xi1> {
  %0 = "tf.LessEqual"(%arg0, %arg1) : (tensor<1xi32>, tensor<1xi32>) -> tensor<1xi1>
  "tf.Assert"(%0, %arg1) {summarize = 3} : (tensor<1xi1>, tensor<1xi32>) -> ()
  func.return %0 : tensor<1xi1>
  // CHECK-LABEL: preserve_assert
  // CHECK: tfl.less_equal
  // CHECK: Assert
  // CHECK: return

  func.return
}

// CHECK-LABEL: @tf_if_gets_control_node
func.func @tf_if_gets_control_node(%arg0: tensor<1xi1>)->() {
 "tf.If"(%arg0) {_lower_using_switch_merge = true, _read_only_resource_inputs = [], device = "", else_branch = @noop, is_stateless = false, output_shapes = [#tf_type.shape<>], then_branch = @noop} : (tensor<1xi1>) -> ()
 func.return
}
// CHECK-NEXT: %[[CONTROL:.*]] = tfl.control_node controls "tf.If"
// CHECK-NEXT: return

	use(a100k[i16])
	use(p1[i16])
	use(p1k[i16])
	use(p100k[i16])

	// Unsigned 16-bit numbers don't need checks for len >= 2¹⁶.
	use(s[ui16])
	use(a1[ui16])
	use(a1k[ui16])
	use(a100k[ui16]) // ERROR "index bounds check elided"
	use(p1[ui16])
	use(p1k[ui16])
	use(p100k[ui16]) // ERROR "index bounds check elided"

	use(s[i32])
	use(a1[i32])
	use(a1k[i32])
	use(a100k[i32])
	use(p1[i32])
	use(p1k[i32])

  %0 = "tfl.less"(%arg0, %arg1) : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xi1>
  %1 = "tf.If"(%0, %arg0, %arg1) {else_branch = @cond_false, then_branch = @cond_true, is_stateless = false} : (tensor<1xi1>, tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
  func.return %1 : tensor<1xf32>
}

    }
  }
}
versions {
  producer: 216
}

# NONE: func @main(%arg0: tensor<i1>, %arg1: tensor<1xi32>) -> (tensor<i1>, tensor<1xi32>)

// CHECK:            shape {
// CHECK-NEXT:         unknown_rank: true


func.func @main() {
  tf_executor.graph {
    %0:4 = tf_executor.island wraps "tf.InfeedDequeueTuple"() : () -> (tensor<3xi32>, tensor<4x?xf32>, tensor<*xi16>)
    tf_executor.fetch
  }
  func.return

  %8 = "tf.opI"() : () -> tensor<*xf32>
  %9 = "tf.opJ"() : () -> tensor<*xi8>
  %10 = "tf.opK"() : () -> tensor<*xi16>
  %11 = "tf.opL"() : () -> tensor<*xi64>
  tf_device.replicate([%0, %1, %2] as %input0: tensor<*xi1>, %9 as %input1: tensor<*xi8>, %10 as %input2: tensor<*xi16>, [%3, %4, %5] as %input3: tensor<*xi32>, [%6, %7, %8] as %input4: tensor<*xf32>, %11 as %input5: tensor<*xi64>) {n = 3 : i32} {
    tf_device.return
  }