Optimize maybe.

[hjchen2]

No description provided.

[PragmaTwice]

template<typename T>
class Maybe<T, typename std::enable_if<std::is_same<T, void>::value>::type> final {
  ...
  SharedOrScalar<cfg::ErrorProto, void*> error_or_scalar_;
  ^^^^^^^^^^^^^^^^^^^^^^^^^
};

这里感觉只需要一个 std::shared_ptr<cfg::ErrorProto>

[hjchen2]

```c++
template<typename T>
class Maybe<T, typename std::enable_if<std::is_same<T, void>::value>::type> final {
  ...
  SharedOrScalar<cfg::ErrorProto, void*> error_or_scalar_;
  ^^^^^^^^^^^^^^^^^^^^^^^^^
};

这里感觉只需要一个 std::shared_ptrcfg::ErrorProto

还有一个is_error_的标记吧，这里就直接复用了SharedOrScalar，也没什么毛病，SharedOrScalar优化之后开销很低了。

[PragmaTwice]

还有一个is_error_的标记吧，这里就直接复用了SharedOrScalar，也没什么毛病，SharedOrScalar优化之后开销很低了。

感觉可以用 error 是不是 null 判断？一般来讲如果有 error 那肯定不能是空的 error。不过确实 SharedOrScalar 也可以

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 139.3ms (= 6964.0ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 127.9ms (= 6394.2ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.09 (= 139.3ms / 127.9ms)

PyTorch resnet50 time: 82.9ms (= 4147.4ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.2ms (= 3712.3ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.12 (= 82.9ms / 74.2ms)

PyTorch resnet50 time: 54.9ms (= 2746.8ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.4ms (= 2371.2ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.16 (= 54.9ms / 47.4ms)

PyTorch resnet50 time: 48.2ms (= 2410.9ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 39.7ms (= 1983.1ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.22 (= 48.2ms / 39.7ms)

PyTorch resnet50 time: 42.5ms (= 2126.8ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 37.9ms (= 1892.5ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.12 (= 42.5ms / 37.9ms)

[chengtbf]

mark。暂时移除 automerge，两个小时以后加回来。

[PragmaTwice]

目前只是跑测试，没有加 automerge，到时候把 bot 的 review 加回来就好

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 141.1ms (= 7053.2ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.1ms (= 6407.5ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 141.1ms / 128.1ms)

PyTorch resnet50 time: 83.8ms (= 4191.6ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.5ms (= 3723.4ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.13 (= 83.8ms / 74.5ms)

PyTorch resnet50 time: 58.5ms (= 2926.4ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 48.2ms (= 2408.9ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.21 (= 58.5ms / 48.2ms)

PyTorch resnet50 time: 48.8ms (= 2439.6ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 41.9ms (= 2092.9ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.17 (= 48.8ms / 41.9ms)

PyTorch resnet50 time: 43.9ms (= 2194.7ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 37.5ms (= 1876.2ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.17 (= 43.9ms / 37.5ms)

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 141.1ms (= 7052.8ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 128.2ms (= 6409.5ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 141.1ms / 128.2ms)

PyTorch resnet50 time: 83.9ms (= 4193.0ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.4ms (= 3718.6ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.13 (= 83.9ms / 74.4ms)

PyTorch resnet50 time: 57.9ms (= 2895.0ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.6ms (= 2378.9ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.22 (= 57.9ms / 47.6ms)

PyTorch resnet50 time: 47.5ms (= 2375.6ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 39.9ms (= 1996.4ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.19 (= 47.5ms / 39.9ms)

PyTorch resnet50 time: 44.1ms (= 2207.4ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 40.6ms (= 2030.8ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.09 (= 44.1ms / 40.6ms)