%1 = "tfl.add"(%0, %cst2) {fused_activation_function = "NONE"} : (tensor<4x4xf32>, tensor<4x4xf32>) -> tensor<4x4xf32>

  func.return %1 : tensor<4x4xf32>

  // CHECK-DAG: %[[bias:.*]] = arith.constant dense<{{.*}}> : tensor<4xf32>
  // CHECK-DAG: %[[filter:.*]] = arith.constant dense<{{.*}}> : tensor<4x4xf32>

  %1 = "tf.AddV2"(%arg0, %0) : (tensor<4x1xf32>, tensor<1x2xf32>) -> tensor<4x2xf32>
  %2 = "tf.AddV2"(%0, %arg0) : (tensor<1x2xf32>, tensor<4x1xf32>) -> tensor<4x2xf32>

  // If operand has the same shape as a result, we can fold it.
  %3 = "tf.AddV2"(%arg1, %0) : (tensor<4x2xf32>, tensor<1x2xf32>) -> tensor<4x2xf32>
  %4 = "tf.AddV2"(%0, %arg1) : (tensor<1x2xf32>, tensor<4x2xf32>) -> tensor<4x2xf32>

  // CHECK: %[[CONST:.*]] = "tf.Const"()

func.func @concat_v2(%arg0: tensor<3x3xf32>, %arg1: tensor<3x3xf32>) -> tensor<6x3xf32> {
  // CHECK: "mhlo.concatenate"({{.*}}) <{dimension = 0 : i64}> : (tensor<3x3xf32>, tensor<3x3xf32>) -> tensor<6x3xf32>
  %axis = "tf.Const"() { value = dense<0> : tensor<i64> } : () -> tensor<i64>
  %1 = "tf.ConcatV2"(%arg0, %arg1, %axis) : (tensor<3x3xf32>, tensor<3x3xf32>, tensor<i64>) -> tensor<6x3xf32>
  func.return %1 : tensor<6x3xf32>
}

    %5 = "quantfork.qcast"(%4) {volatile} : (tensor<1x3xf32>) -> tensor<1x3x!quant.uniform<i8:f32, 0.13170163023705575:-1>>
    %6 = "quantfork.dcast"(%5) : (tensor<1x3x!quant.uniform<i8:f32, 0.13170163023705575:-1>>) -> tensor<1x3xf32>
    %7 = stablehlo.reshape %6 : (tensor<1x3xf32>) -> tensor<3x1xf32>

  // CHECK-DAG: %[[EXP:.*]] = "tf.Exp"(%[[SHIFTED]]) : (tensor<2x3xf32>) -> tensor<2x3xf32>
  // CHECK-DAG: %[[SUM:.*]] = "tf.Sum"(%[[EXP]], %[[AXIS]]) <{keep_dims = true}> : (tensor<2x3xf32>, tensor<1xi64>) -> tensor<2x1xf32>
  // CHECK-DAG: %[[RESULT:.*]] = "tf.Div"(%[[EXP]], %[[SUM]]) : (tensor<2x3xf32>, tensor<2x1xf32>) -> tensor<2x3xf32>
  // CHECK: return %[[RESULT]]

  func.return %24 : tensor<1x4xf32>
}

      %2 = "quantfork.stats"(%1) {layerStats = dense<[5.00000000e-6, 7.00000000e-1]> : tensor<2xf32>} : (tensor<1x3xf32>) -> tensor<1x3xf32>
      return %2 : tensor<1x3xf32>
    }

    func.func private @composite_dot_general_fn(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {

  // CHECK: return %[[MATMUL_1]] : tensor<4x6xf32>
}

// -----

func.func @batchMatMulMatrixAdjXY(%arg0: tensor<5x4xf32>, %arg1: tensor<6x5xf32>) -> tensor<4x6xf32> {
  %0 = "tf.BatchMatMul"(%arg0, %arg1) {adj_x = true, adj_y = true} : (tensor<5x4xf32>, tensor<6x5xf32>) -> tensor<4x6xf32>
  func.return %0 : tensor<4x6xf32>

  // CHECK: device = "cpu"
  %2 = "tf.Relu"(%1) {T = f32, _output_shapes = ["tfshape$dim { size: 3 } dim { size: 3 }"], device = "cpu"} : (tensor<3x3xf32>) -> tensor<3x3xf32>
  // CHECK: device = "xpu"
  %3 = "tf.Relu"(%2) {T = f32, _output_shapes = ["tfshape$dim { size: 3 } dim { size: 3 }"]} : (tensor<3x3xf32>) -> tensor<3x3xf32>
  func.return %3 : tensor<3x3xf32>
}

    version = 5 : i64
  } : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
  return %0 : tensor<1x3xf32>
}

// CHECK-LABEL: func.func @qdq_for_dot_general_weight_empty
// CHECK-SAME: (%[[ARG_0:.+]]: tensor<1x2xf32>) -> tensor<1x3xf32>
// CHECK: %[[CST:.+]] = "tf.Const"() <{value = dense<3.000000e-01> : tensor<2x3xf32>}> : () -> tensor<2x3xf32>