add cpu nms kernel

[BBuf]

sam模型需求

[github-actions]

Code got formatted by CI. Please request CI again if you still want to have this PR merged. If the PR is from a forked repo, please download the patch files from the GitHub Actions web page and apply them locally.

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.8ms (= 4376.0ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.5ms (= 5754.1ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.31 (= 57.5ms / 43.8ms)

OneFlow resnet50 time: 26.5ms (= 2652.6ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 37.6ms (= 3761.0ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.42 (= 37.6ms / 26.5ms)

OneFlow resnet50 time: 19.7ms (= 3945.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 36.0ms (= 7195.8ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.82 (= 36.0ms / 19.7ms)

OneFlow resnet50 time: 18.4ms (= 3670.0ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 31.5ms (= 6307.9ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.72 (= 31.5ms / 18.4ms)

OneFlow resnet50 time: 18.3ms (= 3650.7ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 29.7ms (= 5938.0ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.63 (= 29.7ms / 18.3ms)

OneFlow swin dataloader time: 0.202s (= 40.360s / 200, num_workers=1)
PyTorch swin dataloader time: 0.129s (= 25.707s / 200, num_workers=1)
Relative speed: 0.637 (= 0.129s / 0.202s)

OneFlow swin dataloader time: 0.055s (= 10.962s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.529s / 200, num_workers=4)
Relative speed: 0.596 (= 0.033s / 0.055s)

OneFlow swin dataloader time: 0.031s (= 6.208s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.325s / 200, num_workers=8)
Relative speed: 0.536 (= 0.017s / 0.031s)

❌ OneFlow resnet50 time: 47.8ms (= 4775.1ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.3ms (= 6434.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.35 (= 64.3ms / 47.8ms)

OneFlow resnet50 time: 31.5ms (= 3147.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 46.1ms (= 4607.5ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.46 (= 46.1ms / 31.5ms)

OneFlow resnet50 time: 24.1ms (= 4815.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 43.5ms (= 8694.9ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.81 (= 43.5ms / 24.1ms)

OneFlow resnet50 time: 22.2ms (= 4434.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 36.5ms (= 7303.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.65 (= 36.5ms / 22.2ms)

OneFlow resnet50 time: 21.5ms (= 4302.2ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 34.5ms (= 6901.4ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.60 (= 34.5ms / 21.5ms)

[github-actions]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/10225/