Rename straighten algo interface

[wyg1997]

根据讨论： https://github.com/Oneflow-Inc/OneTeam/issues/1509#issuecomment-1166820859 拉直算法默认并闭，并重命名接口： disable_straighten_algorithm -> enable_straighten_algorithm

[Yipeng1994]

Very good

[strint]

这个名字最好再精确一点，“拉直算法“作为代号ok，enable_dfs_op_schedule 是不是会更好一点?

[Yipeng1994]

严格来说拉直不是dfs，

拉直算法不是纯dps，每有一次传输拉直就会从现有计算里面找节点来执行，这时候跑的是已经出现的节点，也就是深度小的。
可以说拉直在传输节点上是bfs，在传输节点间是dfs。
而且5_3_5只是拉直的其中一种模式，也可以选择纯粹的先进先出模式排序比如2_0_1,这时候也是bfs

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 129.2ms (= 12916.8ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 144.2ms (= 14417.8ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.12 (= 144.2ms / 129.2ms)

OneFlow resnet50 time: 75.7ms (= 7574.4ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 85.1ms (= 8514.7ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.12 (= 85.1ms / 75.7ms)

OneFlow resnet50 time: 49.9ms (= 9984.9ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 62.2ms (= 12433.8ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.25 (= 62.2ms / 49.9ms)

OneFlow resnet50 time: 38.1ms (= 7614.0ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 41.6ms (= 8311.2ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.09 (= 41.6ms / 38.1ms)

OneFlow resnet50 time: 37.2ms (= 7437.1ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 38.2ms (= 7639.6ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.03 (= 38.2ms / 37.2ms)

OneFlow swin dataloader time: 0.260s (= 52.070s / 200, num_workers=1)
PyTorch swin dataloader time: 0.147s (= 29.495s / 200, num_workers=1)
Relative speed: 0.566 (= 0.147s / 0.260s)

OneFlow swin dataloader time: 0.074s (= 14.720s / 200, num_workers=4)
PyTorch swin dataloader time: 0.041s (= 8.203s / 200, num_workers=4)
Relative speed: 0.557 (= 0.041s / 0.074s)

OneFlow swin dataloader time: 0.039s (= 7.814s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.446s / 200, num_workers=8)
Relative speed: 0.569 (= 0.022s / 0.039s)

❌ OneFlow resnet50 time: 144.9ms (= 14486.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 168.5ms (= 16850.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.16 (= 168.5ms / 144.9ms)

OneFlow resnet50 time: 93.9ms (= 9387.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 119.3ms (= 11932.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.27 (= 119.3ms / 93.9ms)

OneFlow resnet50 time: 68.9ms (= 13775.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 89.4ms (= 17885.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.30 (= 89.4ms / 68.9ms)

OneFlow resnet50 time: 56.3ms (= 11265.5ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 77.4ms (= 15486.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.37 (= 77.4ms / 56.3ms)

OneFlow resnet50 time: 51.5ms (= 10309.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 79.3ms (= 15857.2ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.54 (= 79.3ms / 51.5ms)

[Flowingsun007]

这个pr合并了，libai那边应该也需要同步更改一下相应的参数：disable_straighten_algorithm -> enable_straighten_algorithm @wyg1997 @CPFLAME

[wyg1997]

disable_straighten_algorithm

libai main分支有用这个接口吗？我好像没搜到？

[CPFLAME]

这个在 libai_bench_merge_main 和 libai_bench 分支下有

应该把

self.config.disable_straighten_algorithm(True)

修改为

self.config.enable_straighten_algorithm(False)

是么?
@Flowingsun007

[wyg1997]

这个在 libai_bench_merge_main 和 libai_bench 分支下有

应该把

self.config.disable_straighten_algorithm(True)

修改为

self.config.enable_straighten_algorithm(False)

self.config.disable_straighten_algorithm(True) 这行可以直接删掉，因为现在默认这个算法是不开启的，在想开启的地方加上 self.config.enable_straighten_algorithm(True) 就可以