%20 = stablehlo.broadcast_in_dim %9, dims = [0, 1, 2] : (tensor<1x1x1xf32>) -> tensor<8x16x4xf32>
    %21 = stablehlo.multiply %19, %20 : tensor<8x16x4xf32>  // * s1 s2
    %22 = call @uniform_quantize_1(%21, %7, %8) : (tensor<8x16x4xf32>, tensor<1x1x1xf32>, tensor<1x1x1xi8>) -> tensor<8x16x4xi8>
    %23 = call @uniform_dequantize(%22, %7, %8) : (tensor<8x16x4xi8>, tensor<1x1x1xf32>, tensor<1x1x1xi8>) -> tensor<8x16x4xf32>

// CHECK:  "tfl.cos"(%arg0) : (tensor<f32>) -> tensor<f32>
}

func.func @elu(%arg0: tensor<11x16xf32>) -> tensor<11x16xf32> {
  %0 = "tf.Elu"(%arg0) : (tensor<11x16xf32>) -> tensor<11x16xf32>
  func.return %0 : tensor<11x16xf32>

// CHECK-LABEL:elu
// CHECK:  "tfl.elu"(%arg0) : (tensor<11x16xf32>) -> tensor<11x16xf32>
}

func.func @expandDims(%arg0: tensor<2x2xf32>, %arg1: tensor<i32>) -> tensor<1x2x2xf32> {

      return %0 : tensor<1x1x4xf32>
    }
  )mlir";

  const OwningOpRef<ModuleOp> module_op =
      ParseModuleOpString(kXlaCallModuleOpWithQuantizationMethodAttr);
  ASSERT_TRUE(module_op);

}

func.func @matmulNoTransposeAOrB(%arg0: tensor<1x1280xf32>, %arg1: tensor<1280x1000xf32>) -> tensor<1x1000xf32> {
  %166 = "tf.MatMul"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", _output_shapes = ["tfshape$dim { size = 1} dim { size = 1000}"], device = "", name = "matmul", transpose_a = false, transpose_b = false} : (tensor<1x1280xf32>, tensor<1280x1000xf32>) -> tensor<1x1000xf32>
  func.return %166 : tensor<1x1000xf32>

  // CHECK-LABEL: matmulNoTransposeAOrB

// CHECK:         }
func.func @broadcast_add(%arg0: tensor<1x1xf32>, %arg1: tensor<1x1000xf32>) -> (tensor<1x1000xf32>, tensor<1x1000xf32>) {
  %0 = "mhlo.broadcast_in_dim"(%arg0) <{broadcast_dimensions = dense<[0, 1]> : tensor<2xi64>}> : (tensor<1x1xf32>) -> tensor<1x1000xf32>
  %1 = mhlo.add %0, %arg1 : tensor<1x1000xf32>
  %2 = mhlo.add %arg1, %0 : tensor<1x1000xf32>

  %custom_2 = "tfl.custom"(%arg0, %dq_w) {custom_code = "CustomTestOp", custom_option = #tfl<const_bytes : "0x">} : (tensor<1x1x1x1xf32>, tensor<1024x1x1x1xf32>) -> tensor<*xf32>
  %custom_3 = "tfl.custom"(%arg0, %dq_w) {custom_code = "CustomTestOp", custom_option = #tfl<const_bytes : "0x">} : (tensor<1x1x1x1xf32>, tensor<1024x1x1x1xf32>) -> tensor<*xf32>

  // CHECK: %[[ADD:.*]] = tfl.add %[[VAR_VAL]], %arg0 {fused_activation_function = "NONE"} : tensor<1x10xf32>
  // CHECK: "tfl.assign_variable"(%[[RESOURCE]], %[[ADD]]) : (tensor<!tf_type.resource<tensor<1x10xf32>>>, tensor<1x10xf32>) -> ()
  // CHECK: %[[RESULT:.*]] = "tfl.read_variable"(%[[RESOURCE]]) : (tensor<!tf_type.resource<tensor<1x10xf32>>>) -> tensor<1x10xf32>

  // CHECK-LABEL: func @parallel_execute_with_tiled_input
  // CHECK-SAME: (%[[ARG_0:[a-z0-9]*]]: tensor<128x10xf32>, %[[ARG_1:[a-z0-9]*]]: tensor<128x10xf32>, %[[ARG_2:[a-z0-9]*]]: tensor<*xi32>, %[[ARG_3:[a-z0-9]*]]: tensor<*xi32>)
  func.func @parallel_execute_with_tiled_input(%arg0: tensor<128x10xf32>, %arg1: tensor<128x10xf32>, %arg2: tensor<*xi32>, %arg3: tensor<*xi32>) -> (tensor<*xi32>, tensor<*xi1>) {
    // CHECK: tf_device.replicate

module {
func.func @tfInplaceUpdate(%arg0: tensor<2x1x2xf32>) -> tensor<2x1x2xf32> {
  %1 = arith.constant dense<1> : tensor<1xi32>
  %2 = arith.constant dense<2.0> : tensor<1x1x2xf32>
  %3 = "tf.InplaceUpdate"(%arg0, %1, %2) {device = ""}
    : (tensor<2x1x2xf32>, tensor<1xi32>, tensor<1x1x2xf32>) -> tensor<2x1x2xf32>
  func.return %3 : tensor<2x1x2xf32>
}
}

//CHECK: module {

// CHECK-SAME:                 %[[ARG_0:.*]]: tensor<1x1x167xf32>
func.func @dot_general_fn(%arg0: tensor<1x1x167xf32>) -> tensor<1x1x64xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<167x64xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [2] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x1x167xf32>, tensor<167x64xf32>) -> tensor<1x1x64xf32>
  return %1 : tensor<1x1x64xf32>
}