Support symbol placement type in functional.

[hjchen2]

No description provided.

[hjchen2]

这里应该是

const auto& v = JUST(detail::cast<std::vector<std::shared_ptr<Symbol<cfg::SbpParallel>>>>(Borrow()));
auto sbp_list = std::make_shared<std::vector<Symbol<cfg::SbpParallel>>>(v->size());
for (int i = 0; i < v->size(); ++i) {
  sbp_list[i] = *(v->at(i));
}
return sbp_list

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 139.5ms (= 6974.0ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 126.0ms (= 6302.3ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.11 (= 139.5ms / 126.0ms)

PyTorch resnet50 time: 84.3ms (= 4212.6ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.1ms (= 3706.6ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 84.3ms / 74.1ms)

PyTorch resnet50 time: 57.5ms (= 2875.4ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 50.2ms (= 2507.7ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.15 (= 57.5ms / 50.2ms)

PyTorch resnet50 time: 48.3ms (= 2417.5ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 48.6ms (= 2427.8ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.00 (= 48.3ms / 48.6ms)

PyTorch resnet50 time: 44.1ms (= 2205.3ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.4ms (= 2371.9ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 0.93 (= 44.1ms / 47.4ms)

[lixinqi]

"Placement"
"Sbp"
"SbpList"

[lixinqi]

"Placement": "const Symbol<ParallelDesc>&",
"Sbp": "const Symbol<cfg::SbpParallel>&",
"SbpList": "const std::vector<Symbol<cfg::SbpParallel>>&",

[lixinqi]

这么做是因为我们的ParallelDesc后边一定会统一命名成Placement。
但是接口处需要和Python对齐，同时和将来很快要做的改动对齐。