forked from mindspore-Ecosystem/mindspore
96 lines
2.8 KiB
Python
96 lines
2.8 KiB
Python
# Copyright 2019 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 math
|
|
from mindspore import context
|
|
import mindspore.nn as nn
|
|
from mindspore.ops import operations as P
|
|
from mindspore import Tensor
|
|
from tests.ut.python.ops.test_math_ops import VirtualLoss
|
|
import mindspore as ms
|
|
from mindspore.common.api import _executor
|
|
from mindspore.ops import composite as C
|
|
|
|
|
|
class NetWithLoss(nn.Cell):
|
|
def __init__(self, network):
|
|
super(NetWithLoss, self).__init__()
|
|
self.loss = VirtualLoss()
|
|
self.network = network
|
|
|
|
def construct(self, x, y, b):
|
|
predict = self.network(x, y, b)
|
|
return self.loss(predict)
|
|
|
|
|
|
class GradWrap(nn.Cell):
|
|
def __init__(self, network):
|
|
super(GradWrap, self).__init__()
|
|
self.network = network
|
|
|
|
def construct(self, x, y, b):
|
|
return C.grad_all(self.network)(x, y, b)
|
|
|
|
|
|
def loop_config(size):
|
|
config_list = []
|
|
num = 1
|
|
split_list = [num]
|
|
for i in range(int(math.log2(size))):
|
|
num = num * 2
|
|
split_list.append(num)
|
|
|
|
for a in split_list:
|
|
for b in split_list:
|
|
if a * b > size:
|
|
continue
|
|
c = int(size / (a * b))
|
|
config_list.append(((a, b), (b, c)))
|
|
|
|
return config_list
|
|
|
|
|
|
# model_parallel test
|
|
def test_two_matmul():
|
|
class Net(nn.Cell):
|
|
def __init__(self, strategy1, strategy2):
|
|
super().__init__()
|
|
self.matmul1 = P.MatMul().set_strategy(strategy1)
|
|
self.matmul2 = P.MatMul().set_strategy(strategy2)
|
|
|
|
def construct(self, x, y, b):
|
|
out = self.matmul1(x, y)
|
|
out = self.matmul2(out, b)
|
|
return out
|
|
|
|
size = 4
|
|
context.set_auto_parallel_context(device_num=size, global_rank=0)
|
|
x = Tensor(np.ones([128, 32]), dtype=ms.float32)
|
|
y = Tensor(np.ones([32, 64]), dtype=ms.float32)
|
|
b = Tensor(np.ones([64, 64]), dtype=ms.float32)
|
|
|
|
config_list = loop_config(size)
|
|
|
|
count = 0
|
|
for strategy1 in config_list:
|
|
for strategy2 in config_list:
|
|
print("=======current config {}=========".format(count))
|
|
print(strategy1, strategy2)
|
|
net = GradWrap(NetWithLoss(Net(strategy1, strategy2)))
|
|
context.set_auto_parallel_context(parallel_mode="semi_auto_parallel")
|
|
_executor.compile(net, x, y, b)
|
|
count = count + 1
|
|
|