Skip to content

Commit f08cc0f

Browse files
miaobytemiaobyte
committed
nn.functional:functional匹配excuter Op
1 parent cbfcebe commit f08cc0f

23 files changed

Lines changed: 376 additions & 369 deletions

‎doc/front/deepxpy.drawio.svg‎

Lines changed: 1 addition & 1 deletion
Loading

‎front/py/deepx/__init__.py‎

Lines changed: 9 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1,11 +1,17 @@
11
from .tensor import Tensor,Shape,Device,DeviceType
2-
from .tensor import zeros, ones, arange, rand, randn, eye
3-
2+
from deepx.nn.functional import full,zeros,ones,arange,rand,randn,eye
3+
from deepx.nn.functional import add,sub,mul,div
4+
from deepx.nn.functional import matmul
45
__all__ = [
56
'Tensor',
67
'Shape',
78
'Device','DeviceType',
8-
'full','zeros', 'ones', 'arange', 'rand', 'randn', 'eye'
9+
#init
10+
'full','zeros', 'ones', 'arange', 'rand', 'randn', 'eye',
11+
#elementwise
12+
"add","sub","mul","div",
13+
#matmul
14+
"matmul",
915
]
1016

1117
# 为了支持 import deepx as dx 的用法

‎front/py/deepx/autograd/__init__.py‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -2,11 +2,11 @@
22
from .node import Node
33
from .nodetype import NodeType
44
from ._datanode import DataNode
5-
5+
from ._opnode import OpNode
66
__all__ = [
77
'Graph',
88
'Node',
99
'NodeType',
1010
'DataNode',
11-
11+
'OpNode',
1212
]

‎front/py/deepx/nn/__init__.py‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,5 @@
1+
from .deepxir import DeepxIR
2+
3+
__all__ = [
4+
"DeepxIR"
5+
]

front/py/deepx/tensor/deepxir.py renamed to front/py/deepx/nn/deepxir.py

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,4 @@
1-
from typing import Any, Tuple, List, Optional
2-
from ..autograd.graph import Graph
1+
from typing import Tuple, List, Optional
32

43
class DeepxIR:
54
def __init__(self,

‎front/py/deepx/nn/functional.py‎

Whitespace-only changes.

‎front/py/deepx/nn/functional/__init__.py‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,12 @@
1+
from .elementwise import add,sub,mul,div
2+
from .new import newtensor
3+
from .print import printtensor
4+
from .matmul import matmul
5+
from .init import full,zeros,ones,arange,rand,randn,eye
6+
__all__ = [
7+
"newtensor",
8+
"printtensor",
9+
"full","zeros","ones","arange","rand","randn","eye",
10+
"add","sub","mul","div",
11+
"matmul",
12+
]

‎front/py/deepx/nn/functional/elementwise.py‎

Lines changed: 198 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,198 @@
1+
from typing import Optional, Union
2+
from deepx.tensor import Tensor
3+
from deepx.autograd.graph import Graph,DataNode,OpNode
4+
from deepx.nn.deepxir import DeepxIR
5+
from deepx.scheduler import send
6+
7+
def _A_B_elementwiseop_C(
8+
a:Tensor,
9+
b: Tensor,
10+
op:str=None,
11+
out:Tensor=None):
12+
opnode = a.graph.add_op(op)
13+
opnode.add_input(a.node)
14+
opnode.add_input(b.node)
15+
out.node.add_input(opnode)
16+
if a.graph.eager:
17+
ir=DeepxIR(op, a.dtype, [a.node.name, b.node.name], [out.node.name])
18+
send(str(ir))
19+
def _A_b_elementwiseop_C(
20+
a:Tensor,
21+
b: Optional[Union[ float, int]] = None,
22+
op:str=None,
23+
out:Tensor=None):
24+
opnode = a.graph.add_op(op)
25+
opnode.add_input(a.node)
26+
varnode=a.graph.add_var("",b)
27+
opnode.add_input(varnode)
28+
out.node.add_input(opnode)
29+
if a.graph.eager:
30+
varir=DeepxIR("argset", a.dtype, [b], [varnode.name])
31+
send(str(varir))
32+
ir=DeepxIR(op+"_scalar", a.dtype, [a.node.name,varnode.name], [out.node.name])
33+
send(str(ir))
34+
#add
35+
OpNode.register("add")
36+
OpNode.register("add_scalar")
37+
38+
def add(
39+
a:Tensor,
40+
b: Optional[Union[Tensor, float, int]] = None,
41+
out:Tensor=None):
42+
if isinstance(b,Tensor):
43+
_A_B_elementwiseop_C(a,b,"add",out)
44+
else:
45+
_A_b_elementwiseop_C(a,b,"add",out)
46+
47+
48+
#sub
49+
OpNode.register("sub")
50+
OpNode.register("sub_scalar")
51+
52+
def sub(
53+
a:Tensor,
54+
b: Optional[Union[Tensor, float, int]] = None,
55+
out:Tensor=None):
56+
if isinstance(b,Tensor):
57+
_A_B_elementwiseop_C(a,b,"sub",out)
58+
else:
59+
_A_b_elementwiseop_C(a,b,"sub",out)
60+
61+
62+
#mul
63+
OpNode.register("mul")
64+
OpNode.register("mul_scalar")
65+
66+
def mul(
67+
a:Tensor,
68+
b: Optional[Union[Tensor, float, int]] = None,
69+
out:Tensor=None):
70+
if isinstance(b,Tensor):
71+
_A_B_elementwiseop_C(a,b,"mul",out)
72+
else:
73+
_A_b_elementwiseop_C(a,b,"mul",out)
74+
75+
76+
#div
77+
OpNode.register("div")
78+
OpNode.register("div_scalar")
79+
80+
def div(
81+
a:Tensor,
82+
b: Optional[Union[Tensor, float, int]] = None,
83+
out:Tensor=None):
84+
if isinstance(b,Tensor):
85+
_A_B_elementwiseop_C(a,b,"div",out)
86+
else:
87+
_A_b_elementwiseop_C(a,b,"div",out)
88+
89+
90+
91+
92+
# OpNode.register("ReLU", 101)
93+
# OpNode.register("Placeholder", 102)
94+
# OpNode.register("Neg", 103)
95+
# NodeType.register("Less", 104)
96+
# NodeType.register("Equal", 105)
97+
# NodeType.register("Sigmoid", 106)
98+
# NodeType.register("Tanh", 107)
99+
# NodeType.register("Reshape", 108)
100+
# NodeType.register("Transpose", 109)
101+
# NodeType.register("Sum", 110)
102+
# NodeType.register("Mean", 111)
103+
104+
# # 操作节点创建函数
105+
# def matmul(a, b, name=None):
106+
# node = OpNode("MatMul", name)
107+
# node.add_input("a", a)
108+
# node.add_input("b", b)
109+
# return node
110+
111+
# def add(a, b, name=None):
112+
# node = OpNode("Add", name)
113+
# node.add_input("a", a)
114+
# node.add_input("b", b)
115+
# return node
116+
117+
# def relu(x, name=None):
118+
# node = OpNode("ReLU", name)
119+
# node.add_input("x", x)
120+
# return node
121+
122+
# def placeholder(name=None, shape=None):
123+
# node = OpNode("Placeholder", name)
124+
# if shape:
125+
# node.set_attr("shape", shape)
126+
# return node
127+
128+
# def neg(x):
129+
# node = OpNode("Neg")
130+
# node.add_input("x", x)
131+
# return node
132+
133+
# def mul(a, b):
134+
# node = OpNode("Mul")
135+
# node.add_input("a", a)
136+
# node.add_input("b", b)
137+
# return node
138+
139+
# def div(a, b):
140+
# node = OpNode("Div")
141+
# node.add_input("a", a)
142+
# node.add_input("b", b)
143+
# return node
144+
145+
# def sub(a, b):
146+
# node = OpNode("Sub")
147+
# node.add_input("a", a)
148+
# node.add_input("b", b)
149+
# return node
150+
151+
# def less(a, b):
152+
# node = OpNode("Less")
153+
# node.add_input("a", a)
154+
# node.add_input("b", b)
155+
# return node
156+
157+
# def equal(a, b):
158+
# node = OpNode("Equal")
159+
# node.add_input("a", a)
160+
# node.add_input("b", b)
161+
# return node
162+
163+
# def sigmoid(x):
164+
# node = OpNode("Sigmoid")
165+
# node.add_input("x", x)
166+
# return node
167+
168+
# def tanh(x):
169+
# node = OpNode("Tanh")
170+
# node.add_input("x", x)
171+
# return node
172+
173+
# def reshape(x, shape):
174+
# node = OpNode("Reshape")
175+
# node.add_input("x", x)
176+
# node.set_attr("shape", shape)
177+
# return node
178+
179+
# def transpose(x, dim0, dim1):
180+
# node = OpNode("Transpose")
181+
# node.add_input("x", x)
182+
# node.set_attr("dim0", dim0)
183+
# node.set_attr("dim1", dim1)
184+
# return node
185+
186+
# def sum(x, dim=None, keepdim=False):
187+
# node = OpNode("Sum")
188+
# node.add_input("x", x)
189+
# node.set_attr("dim", dim)
190+
# node.set_attr("keepdim", keepdim)
191+
# return node
192+
193+
# def mean(x, dim=None, keepdim=False):
194+
# node = OpNode("Mean")
195+
# node.add_input("x", x)
196+
# node.set_attr("dim", dim)
197+
# node.set_attr("keepdim", keepdim)
198+
# return node

‎front/py/deepx/nn/functional/init.py‎

Lines changed: 41 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,41 @@
1+
from typing import Optional
2+
3+
from deepx.tensor import Tensor
4+
from deepx.nn.deepxir import DeepxIR
5+
from deepx.scheduler import send
6+
7+
def full(*shape, fill_value=0, dtype=None, device=None):
8+
if len(shape) == 1 and isinstance(shape[0], (tuple, list)):
9+
shape = shape[0]
10+
t=Tensor(data=None, shape=shape, dtype=dtype, device=device)
11+
if t.graph.eager:
12+
ir=DeepxIR("constant", t.dtype, [fill_value], [t.node.name])
13+
send(str(ir))
14+
return t
15+
16+
def zeros(*shape, dtype=None, device=None):
17+
return full(*shape, fill_value=0, dtype=dtype, device=device)
18+
19+
def ones(*size, dtype=None, device=None):
20+
return full(*size, fill_value=1, dtype=dtype, device=device)
21+
22+
def rand(*size, dtype=None, device=None):
23+
#TODO
24+
pass
25+
26+
def randn(*size, dtype=None, device=None):
27+
#TODO
28+
pass
29+
30+
def arange(start, end=None, step=1, dtype=None, device=None):
31+
#TODO
32+
pass
33+
34+
def eye(
35+
n:int,
36+
m:Optional[int]=None,
37+
dtype:Optional[str]=None,
38+
device:Optional[str]=None):
39+
#TODO
40+
pass
41+

‎front/py/deepx/nn/functional/matmul.py‎

Lines changed: 19 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,19 @@
1+
from deepx.tensor import Tensor
2+
from deepx.autograd import OpNode
3+
from deepx.nn import DeepxIR
4+
from deepx.scheduler import send
5+
6+
7+
OpNode.register("matmul")
8+
9+
def matmul(
10+
a:Tensor,
11+
b: Tensor,
12+
out:Tensor=None):
13+
opnode = a.graph.add_op("matmul")
14+
opnode.add_input(a.node)
15+
opnode.add_input(b.node)
16+
out.node.add_input(opnode)
17+
if a.graph.eager:
18+
ir=DeepxIR("matmul", a.dtype, [a.node.name,b.node.name], [out.node.name])
19+
send(str(ir))

0 commit comments

Comments
 (0)