// CHECK-DAG: %[[CST_0:.+]] = mhlo.constant dense<[1.000000e-01, 2.000000e-01]> : tensor<2xf32>
  // CHECK-DAG: %[[CST_1:.+]] = mhlo.constant dense<[3.000000e-01, 4.000000e-01]> : tensor<2xf32>
  // CHECK: %[[CST_BCAST:.+]] = "mhlo.broadcast_in_dim"(%[[CST_0]]) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<2xf32>) -> tensor<1x1x3x2xf32>

// TF-DAG: %[[cc_0:.*]] = "tf.ConcatV2"(%[[pc_1]], %[[pc_5]], %[[cst_0]]) : (tensor<2x2xi8>, tensor<2x2xi8>, tensor<i32>) -> tensor<2x4xi8>
// TF-DAG: %[[pc_7:.*]] = "tf.PartitionedCall"(%[[cc_0]], %[[cst_6]], %[[cst_5]]) <{config = "", config_proto = "", executor_type = "", f = @dequantize_i8}
// TF-DAG: return %[[pc_7]]

// XLA-LABEL: func @matmul2_concat_with_int_per_layer
// XLA-DAG: %[[cst_0:.*]] = "tf.Const"() <{value = dense<-1> : tensor<i32>}

// CHECK-LABEL: fuseMulIntoPerTensorConv2dWithQDQs
func.func @fuseMulIntoPerTensorConv2dWithQDQs(%arg0: tensor<256x32x32x3xf32>) -> tensor<256x8x7x3xf32> {
  %cst = arith.constant dense<1.5> : tensor<3xf32>
  %cst_0 = arith.constant dense<[1.0, 2.0, 3.0]> : tensor<3xf32>
  %w = arith.constant dense<2.0> : tensor<3x3x3x3xf32>

// CHECK-DAG:     }
func.func @increase_3rd_operand_3_times() -> tensor<i32> {
  %cst_0 = "arith.constant" () {value = dense<1.0> : tensor<f32>} : () -> tensor<f32>
  %cst_1 = "arith.constant" () {value = dense<0.0> : tensor<f32>} : () -> tensor<f32>
  %cst_2 = "arith.constant" () {value = dense<0> : tensor<i32>} : () -> tensor<i32>
  %0:3 = "tfl.while"(%cst_0, %cst_1, %cst_2) (
    // cond
    {

      "tfl.yield"(%3) : (tensor<i1>) -> ()
  },  {
    ^bb0(%arg3: tensor<*xi32>, %arg4: tensor<*x!tf_type.resource<tensor<*xi32>>>):
      %cst1 = arith.constant dense<2> : tensor<4xi32>
      "tf.AssignAddVariableOp"(%arg4, %cst1) {} : (tensor<*x!tf_type.resource<tensor<*xi32>>>, tensor<4xi32>) -> ()
      %4 = "tf.ReadVariableOp"(%arg4) {dtype = i32} : (tensor<*x!tf_type.resource<tensor<*xi32>>>) -> tensor<*xi32>

// -----

module {
  func.func @matmul(%arg0: tensor<1x2x2x3xf32>) -> (tensor<*xf32>) {
    %cst_0 = "tf.Const"() {value = dense<0.000000e+00> : tensor<2x1024xf32>} : () -> tensor<2x1024xf32>

    %cst_0 = "tf.Const"() {value = dense<[-2.000000e+00, 3.000000e+00]> : tensor<2xf32>} : () -> tensor<2xf32>
    %0 = "quantfork.qcast"(%cst) : (tensor<2x2x3x2xf32>) -> tensor<2x2x3x2x!quant.uniform<i8<-127:127>:f32:3, {4.000000e-03,5.000000e-03}>>

    %cst = "tf.Const"() <{value = dense<0> : tensor<i32>}> : () -> tensor<i32>
    %11:4 = "tf.Split"(%cst, %0) {num_split = 4 : i32} : (tensor<i32>, tensor<128x1024xf32>) -> (tensor<32x1024xf32>, tensor<32x1024xf32>, tensor<32x1024xf32>, tensor<32x1024xf32>)
    %cst_0 = "tf.Const"() <{value = dense<0> : tensor<i32>}> : () -> tensor<i32>

  func.func @init_all_tables()
  attributes {tf_saved_model.exported_names = ["__tf_saved_model_session_initializer"]} {
    %cst = arith.constant dense<[1, 2, 3, 4]> : tensor<4xi64>
    %cst_0 = arith.constant dense<["a", "b", "c", "d"]> : tensor<4x!tf_type.string>

		var m8, m9, m10 map[bool]struct {
		}
		pi64_1 = pi64_0
		m6[func(G3, G2) G1 {
			return (G1(35) | i8_0) << i8_1 / i8_1 &^ i8_1 / i8_1
		}(G3("YBiKg"), G2(122.6))] = make(chan bool)
		ast0 = ast0
		i8_1 = (((G1(10)+i8_1)&i8_0+i8_0)&i8_0&i8_1 ^ i8_1) & i8_1
		am4 = am3
		i32_1 = int32(10) &^ i32_0
		m8[func(float64, G3) bool {
			return func(rune, int16) bool {