Refactor scatter, clamp and pow in cpp instead of in python

[doombeaker]

No description provided.

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 138.8ms (= 6940.5ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 127.9ms (= 6392.8ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.09 (= 138.8ms / 127.9ms)

PyTorch resnet50 time: 84.9ms (= 4244.1ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 74.2ms (= 3711.4ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 84.9ms / 74.2ms)

PyTorch resnet50 time: 57.5ms (= 2873.8ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 47.4ms (= 2367.8ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.21 (= 57.5ms / 47.4ms)

PyTorch resnet50 time: 47.8ms (= 2389.4ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 43.4ms (= 2167.7ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 47.8ms / 43.4ms)

PyTorch resnet50 time: 47.8ms (= 2389.2ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 43.7ms (= 2185.3ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 1.09 (= 47.8ms / 43.7ms)

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

PyTorch resnet50 time: 138.2ms (= 6912.0ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 125.8ms (= 6289.7ms / 50, input_shape=[16, 3, 224, 224], backward is enabled)
Relative speed: 1.10 (= 138.2ms / 125.8ms)

PyTorch resnet50 time: 82.2ms (= 4112.1ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 72.7ms (= 3636.8ms / 50, input_shape=[8, 3, 224, 224], backward is enabled)
Relative speed: 1.13 (= 82.2ms / 72.7ms)

PyTorch resnet50 time: 57.4ms (= 2867.6ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 50.3ms (= 2513.4ms / 50, input_shape=[4, 3, 224, 224], backward is enabled)
Relative speed: 1.14 (= 57.4ms / 50.3ms)

PyTorch resnet50 time: 46.2ms (= 2309.9ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 41.8ms (= 2089.5ms / 50, input_shape=[2, 3, 224, 224], backward is enabled)
Relative speed: 1.11 (= 46.2ms / 41.8ms)

PyTorch resnet50 time: 43.8ms (= 2191.6ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
OneFlow resnet50 time: 49.6ms (= 2478.1ms / 50, input_shape=[1, 3, 224, 224], backward is enabled)
Relative speed: 0.88 (= 43.8ms / 49.6ms)