// We cannot fuse this tfl.mul into the preceding conv op because %cst2 is not broadcast-compatible to %cst0.
  %1 = "tfl.mul"(%0, %cst2) {fused_activation_function = "RELU6"} : (tensor<1x4x4x2xf32>, tensor<4x2xf32>) -> tensor<1x4x4x2xf32>

  func.return %1 : tensor<1x4x4x2xf32>

    // CHECK-NEXT: %[[CMP:.*]] = mhlo.compare GT, %[[INP]], %[[ZERO]], NOTYPE : (tensor<1x4x4x3xf32>, tensor<1x4x4x3xf32>) -> tensor<1x4x4x3xi1>
    // CHECK-NEXT: %[[RES:.*]] = mhlo.select %[[CMP]], %[[INP]], %[[LEAKY]] : tensor<1x4x4x3xi1>, tensor<1x4x4x3xf32>
    // CHECK-NEXT: return %[[RES]] : tensor<1x4x4x3xf32>
    %0 = "tf.LeakyRelu"(%arg0) {alpha = 2.000000e-01 : f32, device = ""} : (tensor<1x4x4x3xf32>) -> tensor<1x4x4x3xf32>

// CHECK:           %[[VAL_2:.*]] = "tf.Conv2D"(%[[VAL_0]], %[[VAL_1]]) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true}> : (tensor<1x6x6x207xf32>, tensor<3x3x207x16xf32>) -> tensor<1x4x4x16xf32>
// CHECK:           return %[[VAL_2]] : tensor<1x4x4x16xf32>
// CHECK:         }