mindspore/tests/ut/python/rewrite/test_net_simple.py

140 lines
4.8 KiB
Python

# Copyright 2022 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
import numpy as np
import mindspore
from mindspore import Tensor, nn
from mindspore.rewrite import SymbolTree, ScopedValue, ValueType, Node
from mindspore.common.initializer import Normal
from mindspore.common.api import _cell_graph_executor
class SimpleNet(nn.Cell):
def __init__(self, num_class=10, num_channel=1):
super(SimpleNet, self).__init__()
self.conv1 = nn.Conv2d(num_channel, 6, 5, pad_mode='valid')
self.conv2 = nn.Conv2d(6, 16, 5, pad_mode='valid')
self.relu = nn.ReLU()
self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
self.flatten = nn.Flatten()
self.fc1 = nn.Dense(16 * 5 * 5, 120, weight_init=Normal(0.02))
self.fc2 = nn.Dense(120, 84, weight_init=Normal(0.02))
self.fc3 = nn.Dense(84, num_class, weight_init=Normal(0.02))
self.var = 10
def construct(self, x):
x = self.conv1(x)
x = x
y = self.var
y = y * 5
y = y and True
x = self.relu(x)
x = self.max_pool2d(x)
x = self.conv2(x)
x = self.relu(x)
x = self.max_pool2d(x)
x = self.flatten(x)
x = self.relu(self.fc1(x))
x = self.relu(self.fc2(x))
x = self.fc3(x)
return x
class MyCell(nn.Cell):
def __init__(self):
super().__init__()
self.conv = nn.Dense(5, 16)
def construct(self, x, y):
x = self.conv(x)
x = mindspore.ops.Add()(x, y)
return x
def add_conv_before_flatten(stree: SymbolTree):
new_conv_node = None
for node in stree.nodes():
if node.get_instance_type() == mindspore.nn.Flatten:
position = stree.before(node)
new_conv = nn.Conv2d(16, 16, 3)
new_conv_node = Node.create_call_cell(new_conv, targets=['x_1'], name='new_conv',
args=[ScopedValue.create_naming_value('self_max_po')])
stree.insert(position, new_conv_node)
break
if new_conv_node is not None:
for node in stree.nodes():
if node.get_instance_type() == mindspore.nn.Flatten:
inputs = node.get_inputs()
assert len(inputs) == 1
new_conv_node.set_arg_by_node(0, inputs[0])
def add_my_cell_after_x_12(stree: SymbolTree):
for node in stree.nodes():
targets = node.get_targets()
if targets is None:
continue
assert targets[0].type == ValueType.NamingValue
target = str(targets[0])
if target == "x_12":
position = stree.after(node)
custom_cell = MyCell()
bias = Tensor(1, mindspore.int32)
new_custom_node = Node.create_call_cell(custom_cell, targets=['nx2'],
args=[ScopedValue.create_naming_value('nx3'),
ScopedValue.create_variable_value(bias)], name='my_cell')
stree.insert(position, new_custom_node)
new_custom_node.set_arg(0, "x_12")
break
def erase_node_x_11(stree: SymbolTree):
return_node = None
for node in stree.nodes():
if node.get_targets() is None:
return_node = node
break
assert return_node is not None
for node in stree.nodes():
targets = node.get_targets()
if targets is None:
continue
assert targets[0].type == ValueType.NamingValue
target = str(targets[0])
if target == "x_11":
stree.set_output(0, "x_10")
stree.erase_node(node)
break
def transform(stree: SymbolTree):
add_conv_before_flatten(stree)
erase_node_x_11(stree)
def test_simple_net():
"""
Feature: Module rewrite.
Description: Resolve a simple network by rewrite and do some transform on it.
Expectation: Result of rewrite can be compiled.
"""
net = SimpleNet(10)
stree = SymbolTree.create(net)
transform(stree)
print("------------------------------------ keys of global_vars: ", stree.get_handler().get_global_vars().keys())
net_opt = stree.get_network()
data_in = Tensor(np.ones([1, 1, 32, 32]), mindspore.float32)
_cell_graph_executor.compile(net_opt, data_in)