Move pybind/python api to cpython [part2]

[marigoold]

本PR完成了：

• 重命名了 move pybind11/Python api to cpython #8258 中as_number的一系列函数和宏，防止其与method中的函数命名冲突。比如重命名+运算符对应的函数 PyTensorObject_add 为 PyTensorObject_nb_add，防止其与 Tensor.add 对应的 PyTensorObject_add 冲突。
• 搬运了比 move pybind11/Python api to cpython #8258 中稍微复杂的部分api

这部分API的搬运思路如下：

• 实现了工具函数 concat_self ，用来把Tensor self和args拼接起来，以便生成的函数接口调用
• 对于参数解析比较复杂的函数，用一个宏 DIRECT_PASS_FUNC 来创建函数，统一调用 functional_api.yaml.pybind.h 里面的接口
• 对于参数解析比较简单的函数，在Python C api层就完成参数解析，然后调用functional接口
• 对于 Tensor.int/float 等转dtype的函数，用宏 DATATYPE_FUNC 创建（虽然只有四个），不调用 functional_api.yaml.pybind.h 里面的接口

其他：

• 对于部分functional_api.yaml中参数名和PyTorch对应函数参数名不对齐的函数，改成了和PyTorch相同的参数名
• cast函数原来没有pin_memory参数，现在增加

这个部分迁移的api如下：

• Tensor.ndim = property(_ndim)
• Tensor.numpy = _numpy
• Tensor.size = _size
• Tensor.backward = _backward
• Tensor.setitem = _setitem
• Tensor.str = _str
• Tensor.repr = _repr
• Tensor.bool = is_nonzero
• Tensor.iadd = _iadd
• Tensor.addmm = _addmm
• Tensor.format = _format
• Tensor.index = _index
• Tensor.float = _scalar_float
• Tensor.int = _scalar_int
• Tensor.array = _numpy
• Tensor.uniform_ = _uniform
• Tensor.trunc_normal_ = trunc_normal
• Tensor.kaiming_uniform_ = _kaiming_uniform
• Tensor.kaiming_normal_ = _kaiming_normal
• Tensor.xavier_normal_ = _xavier_normal
• Tensor.xavier_uniform_ = _xavier_uniform
• Tensor.orthogonal_ = _orthogonal
• Tensor.normal_ = _normal
• Tensor.fill_ = _fill
• Tensor.copy_ = _copy
• Tensor._meta_repr = _meta_repr
• Tensor.floor_divide = _floor_divide
• Tensor.argmax = _argmax
• Tensor.argmin = _argmin
• Tensor.argsort = _argsort
• Tensor.argwhere = _argwhere
• Tensor.amin = _amin
• Tensor.atan2 = _atan2
• Tensor.gt = _gt
• Tensor.ge = _ge
• Tensor.cast = _cast
• Tensor.diag = _diag
• Tensor.diagonal = _diagonal
• Tensor.add = _add
• Tensor.add_ = _add_inplace
• Tensor.addcmul = _addcmul
• Tensor.addcmul_ = addcmul
• Tensor.div = _truediv
• Tensor.div_ = _truediv_inplace
• Tensor.mul = _mul
• Tensor.mul_ = mul
• Tensor.sub = _sub
• Tensor.sub_ = _sub_inplace
• Tensor.clamp = _clamp
• Tensor.clamp_ = clamp
• Tensor.clip = _clip
• Tensor.clip_ = clip
• Tensor.cpu = _cpu
• Tensor.cuda = _cuda
• Tensor.expand = _expand
• Tensor.expand_as = _expand_as
• Tensor.fmod = _fmod
• Tensor.flatten = _flatten
• Tensor.flip = _flip
• Tensor.in_top_k = _in_top_k
• Tensor.index_select = _index_select
• Tensor.minimum = _minimum
• Tensor.maximum = _maximum
• Tensor.new_empty = _new_empty
• Tensor.new_ones = _new_ones
• Tensor.new_zeros = _new_zeros
• Tensor.pow = _pow
• Tensor.var = _var
• Tensor.std = _std
• Tensor.matmul = _matmul
• Tensor.softplus = _softplus
• Tensor.tril = _tril
• Tensor.triu = _triu
• Tensor.where = _where
• Tensor.norm = _norm
• Tensor.transpose = _transpose
• Tensor.permute = _permute
• Tensor.local_to_global = _local_to_global
• Tensor.global_to_global = _global_to_global
• Tensor.to_global = _to_global
• Tensor.relu = _relu
• Tensor.relu_ = _relu_inplace
• Tensor.softmax = _softmax
• Tensor.log_softmax = _log_softmax
• Tensor.logical_and = _and
• Tensor.logical_or = _or
• Tensor.logical_not = _not
• Tensor.logical_xor = _xor
• Tensor.roll = _roll
• Tensor.bmm = _bmm
• Tensor.chunk = _chunk
• Tensor.repeat = _repeat
• Tensor.tile = _tile
• Tensor.split = _split
• Tensor.unbind = _unbind
• Tensor.squeeze = _squeeze
• Tensor.swapaxes = _swapaxes
• Tensor.amax = _amax
• Tensor.swapdims = _swapdims
• Tensor.unfold = _unfold
• Tensor.narrow = _narrow
• Tensor.unsqueeze = _unsqueeze
• Tensor.to = _to
• Tensor.half = _half
• Tensor.gather = _gather
• Tensor.all = _all
• Tensor.any = _any
• Tensor.T = property(_T)
• Tensor.masked_fill = _masked_fill
• Tensor.masked_select = _masked_select
• Tensor.eq = _eq
• Tensor.ne = _ne
• Tensor.item = _item
• Tensor.lt = _lt
• Tensor.le = _le
• Tensor.to_local = _to_local
• Tensor.reshape = _reshape
• Tensor.reshape_as = _reshape_as
• Tensor.view = _view
• Tensor.view_as = _view_as
• Tensor.sort = _sort
• Tensor.type_as = _type_as
• Tensor.tolist = _tolist
• Tensor.int = _int
• Tensor.long = _long
• Tensor.float = _float
• Tensor.double = _double
• Tensor.is_floating_point = _is_floating_point
• Tensor.topk = _topk
• Tensor.nms = _nms
• Tensor.nonzero = _nonzero
• Tensor.max = _max
• Tensor.min = _min
• Tensor.median = _median
• Tensor.sum = _sum
• Tensor.mean = _mean
• Tensor.prod = _prod
• Tensor.is_consistent = _is_consistent
• Tensor.to_consistent = _to_consistent
• Tensor.new_tensor = _new_tensor
• Tensor.cumsum = _cumsum
• Tensor.cumprod = _cumprod

[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]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8315/

[github-actions]

Speed stats:

GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 130.0ms (= 13004.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 143.7ms (= 14371.9ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.11 (= 143.7ms / 130.0ms)

OneFlow resnet50 time: 80.4ms (= 8042.2ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 84.9ms (= 8489.5ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.06 (= 84.9ms / 80.4ms)

OneFlow resnet50 time: 53.7ms (= 10749.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 57.8ms (= 11553.1ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.07 (= 57.8ms / 53.7ms)

OneFlow resnet50 time: 42.9ms (= 8575.6ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 45.9ms (= 9186.2ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.07 (= 45.9ms / 42.9ms)

OneFlow resnet50 time: 36.3ms (= 7255.3ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.7ms (= 8131.1ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.12 (= 40.7ms / 36.3ms)

OneFlow swin dataloader time: 0.249s (= 49.819s / 200, num_workers=1)
PyTorch swin dataloader time: 0.151s (= 30.288s / 200, num_workers=1)
Relative speed: 0.608 (= 0.151s / 0.249s)

OneFlow swin dataloader time: 0.065s (= 13.022s / 200, num_workers=4)
PyTorch swin dataloader time: 0.042s (= 8.367s / 200, num_workers=4)
Relative speed: 0.643 (= 0.042s / 0.065s)

OneFlow swin dataloader time: 0.036s (= 7.198s / 200, num_workers=8)
PyTorch swin dataloader time: 0.023s (= 4.575s / 200, num_workers=8)
Relative speed: 0.636 (= 0.023s / 0.036s)

❌ OneFlow resnet50 time: 146.7ms (= 14675.0ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 172.3ms (= 17226.1ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.17 (= 172.3ms / 146.7ms)

OneFlow resnet50 time: 96.3ms (= 9634.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 112.7ms (= 11266.4ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.17 (= 112.7ms / 96.3ms)

OneFlow resnet50 time: 72.7ms (= 14543.4ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 89.8ms (= 17950.9ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.23 (= 89.8ms / 72.7ms)

OneFlow resnet50 time: 57.2ms (= 11446.4ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 74.1ms (= 14821.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.29 (= 74.1ms / 57.2ms)

OneFlow resnet50 time: 54.9ms (= 10988.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 69.5ms (= 13905.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.27 (= 69.5ms / 54.9ms)

[Flowingsun007]

shape->NumAxes()用了很多次，建议提取出来避免重复调用

[marigoold]

已改

[hjchen2]

同时也要把这里不用的接口也删一下，比如_double、_float这些