Skip to content
体验新版
项目
组织
正在加载...
登录
切换导航
打开侧边栏
PaddlePaddle
Paddle
提交
bd018360
P
Paddle
项目概览
PaddlePaddle
/
Paddle
接近 1 年 前同步成功
通知
2292
Star
20931
Fork
5422
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
1423
列表
看板
标记
里程碑
合并请求
543
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
P
Paddle
项目概览
项目概览
详情
发布
仓库
仓库
文件
提交
分支
标签
贡献者
分支图
比较
Issue
1,423
Issue
1,423
列表
看板
标记
里程碑
合并请求
543
合并请求
543
Pages
分析
分析
仓库分析
DevOps
Wiki
0
Wiki
成员
成员
收起侧边栏
关闭侧边栏
动态
分支图
创建新Issue
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
未验证
提交
bd018360
编写于
6月 01, 2022
作者:
C
caozhou
提交者:
GitHub
6月 01, 2022
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
add some comp op costs (#43114)
上级
010aba33
变更
2
展开全部
隐藏空白更改
内联
并排
Showing
2 changed file
with
752 addition
and
36 deletion
+752
-36
python/paddle/distributed/auto_parallel/cost/comp_op_cost.py
python/paddle/distributed/auto_parallel/cost/comp_op_cost.py
+574
-36
python/paddle/fluid/tests/unittests/auto_parallel/test_comp_cost.py
...dle/fluid/tests/unittests/auto_parallel/test_comp_cost.py
+178
-0
未找到文件。
python/paddle/distributed/auto_parallel/cost/comp_op_cost.py
浏览文件 @
bd018360
此差异已折叠。
点击以展开。
python/paddle/fluid/tests/unittests/auto_parallel/test_comp_cost.py
浏览文件 @
bd018360
...
...
@@ -54,6 +54,35 @@ from paddle.distributed.auto_parallel.cost.comp_op_cost import LookupTableV2Grad
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
MatmulOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
MatmulGradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
MatmulV2OpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
MatmulV2GradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
MemcpyOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
MulOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
MulGradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
OneHotOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
ReadFromArrayOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
ReduceSumOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
ReduceSumGradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
Reshape2OpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
Reshape2GradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
ReduceMeanOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
ReduceMeanGradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SamplingIdOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
ScaleOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SliceOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SoftmaxOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SoftmaxGradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SoftmaxWithCrossEntropyOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SoftmaxWithCrossEntropyGradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SplitOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
Squeeze2OpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SquareOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SquareGradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
SumOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
TopKOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
Transpose2OpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
Transpose2GradOpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
Unsqueeze2OpCost
from
paddle.distributed.auto_parallel.cost.comp_op_cost
import
WriteToArrayOpCost
from
test_cluster
import
cluster_json
...
...
@@ -244,6 +273,155 @@ class TestCompOpCost(unittest.TestCase):
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
MatmulV2GradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
MemcpyOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
MulOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
MulGradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
OneHotOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
ReadFromArrayOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
ReduceSumOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
ReduceSumGradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
Reshape2OpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
MatmulV2OpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
Reshape2GradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
ReduceMeanOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
ReduceMeanGradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SamplingIdOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
ScaleOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SliceOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SoftmaxOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SoftmaxGradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SoftmaxWithCrossEntropyOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SoftmaxWithCrossEntropyGradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SplitOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
Squeeze2OpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SquareOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SquareGradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
SumOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
TopKOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
Transpose2OpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
Transpose2GradOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
Unsqueeze2OpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
op_cost
=
WriteToArrayOpCost
(
cluster
=
cluster
)
self
.
assertTrue
(
op_cost
.
flops
>=
0
)
self
.
assertTrue
(
op_cost
.
time
>=
0
)
self
.
assertTrue
(
op_cost
.
memory
>=
0
)
# Remove unnecessary files
if
os
.
path
.
exists
(
cluster_json_path
):
os
.
remove
(
cluster_json_path
)
...
...
编辑
预览
Markdown
is supported
0%
请重试
或
添加新附件
.
添加附件
取消
You are about to add
0
people
to the discussion. Proceed with caution.
先完成此消息的编辑!
取消
想要评论请
注册
或
登录