Align Torch.nn.functional poolXd

[MARD1NO]

对齐torch.nn.functional.pool系列算子

[github-actions]

Speed stats:

GPU Name: GeForce GTX 1080 

OneFlow resnet50 time: 127.9ms (= 6395.0ms / 50, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 142.3ms (= 7114.9ms / 50, input_shape=[16, 3, 224, 224])
Relative speed: 1.11 (= 142.3ms / 127.9ms)

OneFlow resnet50 time: 74.4ms (= 3718.9ms / 50, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 86.2ms (= 4309.0ms / 50, input_shape=[8, 3, 224, 224])
Relative speed: 1.16 (= 86.2ms / 74.4ms)

OneFlow resnet50 time: 48.0ms (= 2398.2ms / 50, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 57.2ms (= 2857.8ms / 50, input_shape=[4, 3, 224, 224])
Relative speed: 1.19 (= 57.2ms / 48.0ms)

OneFlow resnet50 time: 43.3ms (= 2167.4ms / 50, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 49.1ms (= 2456.0ms / 50, input_shape=[2, 3, 224, 224])
Relative speed: 1.13 (= 49.1ms / 43.3ms)

OneFlow resnet50 time: 41.0ms (= 2050.2ms / 50, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 41.5ms (= 2073.8ms / 50, input_shape=[1, 3, 224, 224])
Relative speed: 1.01 (= 41.5ms / 41.0ms)

OneFlow resnet50 time: 157.7ms (= 7886.6ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 161.6ms (= 8081.2ms / 50, input_shape=[16, 3, 224, 224], ddp, world size=2)
Relative speed: 1.02 (= 161.6ms / 157.7ms)

OneFlow resnet50 time: 104.1ms (= 5207.3ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 102.8ms (= 5142.4ms / 50, input_shape=[8, 3, 224, 224], ddp, world size=2)
Relative speed: 0.99 (= 102.8ms / 104.1ms)

OneFlow resnet50 time: 77.8ms (= 3889.3ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 73.9ms (= 3697.1ms / 50, input_shape=[4, 3, 224, 224], ddp, world size=2)
Relative speed: 0.95 (= 73.9ms / 77.8ms)

OneFlow resnet50 time: 69.0ms (= 3450.6ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.2ms (= 3210.2ms / 50, input_shape=[2, 3, 224, 224], ddp, world size=2)
Relative speed: 0.93 (= 64.2ms / 69.0ms)

OneFlow resnet50 time: 71.5ms (= 3572.6ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.3ms (= 3264.8ms / 50, input_shape=[1, 3, 224, 224], ddp, world size=2)
Relative speed: 0.91 (= 65.3ms / 71.5ms)