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
Xiaomi
Mace
提交
1fdb5593
Mace
项目概览
Xiaomi
/
Mace
通知
106
Star
40
Fork
27
代码
文件
提交
分支
Tags
贡献者
分支图
Diff
Issue
0
列表
看板
标记
里程碑
合并请求
0
DevOps
流水线
流水线任务
计划
Wiki
0
Wiki
分析
仓库
DevOps
项目成员
Pages
Mace
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
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
Pages
DevOps
DevOps
流水线
流水线任务
计划
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
流水线任务
提交
Issue看板
体验新版 GitCode,发现更多精彩内容 >>
提交
1fdb5593
编写于
4月 04, 2018
作者:
李
李寅
浏览文件
操作
浏览文件
下载
差异文件
Merge branch 'master' into 'master'
fix throughput_test to be able to run Closes #175 See merge request !352
上级
fb84272e
eb7589a9
变更
2
显示空白变更内容
内联
并排
Showing
2 changed file
with
122 addition
and
38 deletion
+122
-38
mace/benchmark/model_throughput_test.cc
mace/benchmark/model_throughput_test.cc
+119
-37
tools/build_run_throughput_test.sh
tools/build_run_throughput_test.sh
+3
-1
未找到文件。
mace/benchmark/model_throughput_test.cc
浏览文件 @
1fdb5593
...
...
@@ -25,6 +25,7 @@
#include "mace/public/mace.h"
#include "mace/utils/env_time.h"
#include "mace/utils/logging.h"
#include "mace/core/types.h"
namespace
mace
{
...
...
@@ -72,10 +73,43 @@ extern const std::string ModelChecksum();
namespace
benchmark
{
void
Split
(
const
std
::
string
&
str
,
char
delims
,
std
::
vector
<
std
::
string
>
*
result
)
{
MACE_CHECK_NOTNULL
(
result
);
std
::
string
tmp
=
str
;
while
(
!
tmp
.
empty
())
{
size_t
next_offset
=
tmp
.
find
(
delims
);
result
->
push_back
(
tmp
.
substr
(
0
,
next_offset
));
if
(
next_offset
==
std
::
string
::
npos
)
{
break
;
}
else
{
tmp
=
tmp
.
substr
(
next_offset
+
1
);
}
}
}
void
SplitAndParseToInts
(
const
std
::
string
&
str
,
char
delims
,
std
::
vector
<
int64_t
>
*
result
)
{
MACE_CHECK_NOTNULL
(
result
);
std
::
string
tmp
=
str
;
while
(
!
tmp
.
empty
())
{
index_t
dim
=
atoi
(
tmp
.
data
());
result
->
push_back
(
dim
);
size_t
next_offset
=
tmp
.
find
(
delims
);
if
(
next_offset
==
std
::
string
::
npos
)
{
break
;
}
else
{
tmp
=
tmp
.
substr
(
next_offset
+
1
);
}
}
}
void
ParseShape
(
const
std
::
string
&
str
,
std
::
vector
<
int64_t
>
*
shape
)
{
std
::
string
tmp
=
str
;
while
(
!
tmp
.
empty
())
{
int
dim
=
atoi
(
tmp
.
data
());
in
dex_
t
dim
=
atoi
(
tmp
.
data
());
shape
->
push_back
(
dim
);
size_t
next_offset
=
tmp
.
find
(
","
);
if
(
next_offset
==
std
::
string
::
npos
)
{
...
...
@@ -86,6 +120,14 @@ void ParseShape(const std::string &str, std::vector<int64_t> *shape) {
}
}
std
::
string
FormatName
(
const
std
::
string
input
)
{
std
::
string
res
=
input
;
for
(
size_t
i
=
0
;
i
<
input
.
size
();
++
i
)
{
if
(
!::
isalnum
(
res
[
i
]))
res
[
i
]
=
'_'
;
}
return
res
;
}
DeviceType
ParseDeviceType
(
const
std
::
string
&
device_str
)
{
if
(
device_str
.
compare
(
"CPU"
)
==
0
)
{
return
DeviceType
::
CPU
;
...
...
@@ -100,6 +142,10 @@ DeviceType ParseDeviceType(const std::string &device_str) {
}
}
DEFINE_string
(
input_node
,
"input_node0,input_node1"
,
"input nodes, separated by comma"
);
DEFINE_string
(
output_node
,
"output_node0,output_node1"
,
"output nodes, separated by comma"
);
DEFINE_string
(
input_shape
,
"1,224,224,3"
,
"input shape, separated by comma"
);
DEFINE_string
(
output_shape
,
"1,224,224,2"
,
"output shape, separated by comma"
);
DEFINE_string
(
input_file
,
""
,
"input file name"
);
...
...
@@ -113,7 +159,6 @@ int Main(int argc, char **argv) {
gflags
::
ParseCommandLineFlags
(
&
argc
,
&
argv
,
true
);
LOG
(
INFO
)
<<
"mace version: "
<<
MaceVersion
();
LOG
(
INFO
)
<<
"mace git version: "
<<
MaceGitVersion
();
#ifdef MACE_CPU_MODEL_TAG
LOG
(
INFO
)
<<
"cpu model checksum: "
<<
mace
::
MACE_CPU_MODEL_TAG
::
ModelChecksum
();
...
...
@@ -126,7 +171,9 @@ int Main(int argc, char **argv) {
LOG
(
INFO
)
<<
"dsp model checksum: "
<<
mace
::
MACE_DSP_MODEL_TAG
::
ModelChecksum
();
#endif
LOG
(
INFO
)
<<
"Input node: ["
<<
FLAGS_input_node
<<
"]"
;
LOG
(
INFO
)
<<
"input_shape: "
<<
FLAGS_input_shape
;
LOG
(
INFO
)
<<
"Output node: ["
<<
FLAGS_output_node
<<
"]"
;
LOG
(
INFO
)
<<
"output_shape: "
<<
FLAGS_output_shape
;
LOG
(
INFO
)
<<
"input_file: "
<<
FLAGS_input_file
;
LOG
(
INFO
)
<<
"cpu_model_data_file: "
<<
FLAGS_cpu_model_data_file
;
...
...
@@ -134,31 +181,63 @@ int Main(int argc, char **argv) {
LOG
(
INFO
)
<<
"dsp_model_data_file: "
<<
FLAGS_dsp_model_data_file
;
LOG
(
INFO
)
<<
"run_seconds: "
<<
FLAGS_run_seconds
;
std
::
vector
<
int64_t
>
input_shape_vec
;
std
::
vector
<
int64_t
>
output_shape_vec
;
ParseShape
(
FLAGS_input_shape
,
&
input_shape_vec
);
ParseShape
(
FLAGS_output_shape
,
&
output_shape_vec
);
std
::
vector
<
std
::
string
>
input_names
;
std
::
vector
<
std
::
string
>
output_names
;
std
::
vector
<
std
::
string
>
input_shapes
;
std
::
vector
<
std
::
string
>
output_shapes
;
Split
(
FLAGS_input_node
,
','
,
&
input_names
);
Split
(
FLAGS_output_node
,
','
,
&
output_names
);
Split
(
FLAGS_input_shape
,
':'
,
&
input_shapes
);
Split
(
FLAGS_output_shape
,
':'
,
&
output_shapes
);
const
size_t
input_count
=
input_shapes
.
size
();
const
size_t
output_count
=
output_shapes
.
size
();
std
::
vector
<
std
::
vector
<
int64_t
>>
input_shape_vec
(
input_count
);
std
::
vector
<
std
::
vector
<
int64_t
>>
output_shape_vec
(
output_count
);
for
(
size_t
i
=
0
;
i
<
input_count
;
++
i
)
{
ParseShape
(
input_shapes
[
i
],
&
input_shape_vec
[
i
]);
}
for
(
size_t
i
=
0
;
i
<
output_count
;
++
i
)
{
ParseShape
(
output_shapes
[
i
],
&
output_shape_vec
[
i
]);
}
std
::
map
<
std
::
string
,
mace
::
MaceTensor
>
inputs
;
std
::
map
<
std
::
string
,
mace
::
MaceTensor
>
cpu_outputs
;
std
::
map
<
std
::
string
,
mace
::
MaceTensor
>
gpu_outputs
;
std
::
map
<
std
::
string
,
mace
::
MaceTensor
>
dsp_outputs
;
for
(
size_t
i
=
0
;
i
<
input_count
;
++
i
)
{
// Allocate input and output
int64_t
input_size
=
std
::
accumulate
(
input_shape_vec
.
begin
(),
input_shape_vec
.
end
(),
1
,
std
::
accumulate
(
input_shape_vec
[
i
].
begin
(),
input_shape_vec
[
i
]
.
end
(),
1
,
std
::
multiplies
<
int64_t
>
());
int64_t
output_size
=
std
::
accumulate
(
output_shape_vec
.
begin
(),
output_shape_vec
.
end
(),
1
,
std
::
multiplies
<
int64_t
>
());
std
::
unique_ptr
<
float
[]
>
input_data
(
new
float
[
input_size
]);
std
::
unique_ptr
<
float
[]
>
cpu_output_data
(
new
float
[
output_size
]);
std
::
unique_ptr
<
float
[]
>
gpu_output_data
(
new
float
[
output_size
]);
std
::
unique_ptr
<
float
[]
>
dsp_output_data
(
new
float
[
output_size
]);
auto
buffer_in
=
std
::
shared_ptr
<
float
>
(
new
float
[
input_size
],
std
::
default_delete
<
float
[]
>
());
// load input
std
::
ifstream
in_file
(
FLAGS_input_file
,
std
::
ios
::
in
|
std
::
ios
::
binary
);
std
::
ifstream
in_file
(
FLAGS_input_file
+
"_"
+
FormatName
(
input_names
[
i
]),
std
::
ios
::
in
|
std
::
ios
::
binary
);
if
(
in_file
.
is_open
())
{
in_file
.
read
(
reinterpret_cast
<
char
*>
(
input_data
.
get
()),
in_file
.
read
(
reinterpret_cast
<
char
*>
(
buffer_in
.
get
()),
input_size
*
sizeof
(
float
));
in_file
.
close
();
}
else
{
LOG
(
INFO
)
<<
"Open input file failed"
;
return
-
1
;
LOG
(
FATAL
)
<<
"Open input file failed"
;
}
inputs
[
input_names
[
i
]]
=
mace
::
MaceTensor
(
input_shape_vec
[
i
],
buffer_in
);
}
for
(
size_t
i
=
0
;
i
<
output_count
;
++
i
)
{
int64_t
output_size
=
std
::
accumulate
(
output_shape_vec
[
i
].
begin
(),
output_shape_vec
[
i
].
end
(),
1
,
std
::
multiplies
<
int64_t
>
());
auto
buffer_out
=
std
::
shared_ptr
<
float
>
(
new
float
[
output_size
],
std
::
default_delete
<
float
[]
>
());
cpu_outputs
[
output_names
[
i
]]
=
mace
::
MaceTensor
(
output_shape_vec
[
i
],
buffer_out
);
gpu_outputs
[
output_names
[
i
]]
=
mace
::
MaceTensor
(
output_shape_vec
[
i
],
buffer_out
);
dsp_outputs
[
output_names
[
i
]]
=
mace
::
MaceTensor
(
output_shape_vec
[
i
],
buffer_out
);
}
int64_t
t0
,
t1
,
init_micros
;
...
...
@@ -170,11 +249,12 @@ int Main(int argc, char **argv) {
FLAGS_cpu_model_data_file
.
c_str
());
NetDef
cpu_net_def
=
mace
::
MACE_CPU_MODEL_TAG
::
CreateNet
(
cpu_model_data
);
mace
::
MaceEngine
cpu_engine
(
&
cpu_net_def
,
DeviceType
::
CPU
);
mace
::
MaceEngine
cpu_engine
(
&
cpu_net_def
,
DeviceType
::
CPU
,
input_names
,
output_names
);
LOG
(
INFO
)
<<
"CPU Warm up run"
;
t0
=
NowMicros
();
cpu_engine
.
Run
(
input
_data
.
get
(),
input_shape_vec
,
cpu_output_data
.
get
()
);
cpu_engine
.
Run
(
input
s
,
&
cpu_outputs
);
t1
=
NowMicros
();
LOG
(
INFO
)
<<
"CPU 1st warm up run latency: "
<<
t1
-
t0
<<
" us"
;
#endif
...
...
@@ -187,12 +267,13 @@ int Main(int argc, char **argv) {
FLAGS_gpu_model_data_file
.
c_str
());
NetDef
gpu_net_def
=
mace
::
MACE_GPU_MODEL_TAG
::
CreateNet
(
gpu_model_data
);
mace
::
MaceEngine
gpu_engine
(
&
gpu_net_def
,
DeviceType
::
OPENCL
);
mace
::
MaceEngine
gpu_engine
(
&
gpu_net_def
,
DeviceType
::
OPENCL
,
input_names
,
output_names
);
mace
::
MACE_GPU_MODEL_TAG
::
UnloadModelData
(
gpu_model_data
);
LOG
(
INFO
)
<<
"GPU Warm up run"
;
t0
=
NowMicros
();
gpu_engine
.
Run
(
input
_data
.
get
(),
input_shape_vec
,
gpu_output_data
.
get
()
);
gpu_engine
.
Run
(
input
s
,
&
gpu_outputs
);
t1
=
NowMicros
();
LOG
(
INFO
)
<<
"GPU 1st warm up run latency: "
<<
t1
-
t0
<<
" us"
;
#endif
...
...
@@ -202,15 +283,16 @@ int Main(int argc, char **argv) {
LOG
(
INFO
)
<<
"Load & init dsp model and warm up"
;
const
unsigned
char
*
dsp_model_data
=
mace
::
MACE_DSP_MODEL_TAG
::
LoadModelData
(
FLAGS_
gpu
_model_data_file
.
c_str
());
FLAGS_
dsp
_model_data_file
.
c_str
());
NetDef
dsp_net_def
=
mace
::
MACE_DSP_MODEL_TAG
::
CreateNet
(
dsp_model_data
);
mace
::
MaceEngine
dsp_engine
(
&
dsp_net_def
,
DeviceType
::
HEXAGON
);
mace
::
MaceEngine
dsp_engine
(
&
dsp_net_def
,
DeviceType
::
HEXAGON
,
input_names
,
output_names
);
mace
::
MACE_DSP_MODEL_TAG
::
UnloadModelData
(
dsp_model_data
);
LOG
(
INFO
)
<<
"DSP Warm up run"
;
t0
=
NowMicros
();
gpu_engine
.
Run
(
input_data
.
get
(),
input_shape_vec
,
dsp_output_data
.
get
()
);
dsp_engine
.
Run
(
inputs
,
&
dsp_outputs
);
t1
=
NowMicros
();
LOG
(
INFO
)
<<
"DSP 1st warm up run latency: "
<<
t1
-
t0
<<
" us"
;
#endif
...
...
@@ -226,7 +308,7 @@ int Main(int argc, char **argv) {
int64_t
micros
=
0
;
int64_t
start
=
NowMicros
();
for
(;
micros
<
run_micros
;
++
frames
)
{
cpu_engine
.
Run
(
input
_data
.
get
(),
input_shape_vec
,
cpu_output_data
.
get
()
);
cpu_engine
.
Run
(
input
s
,
&
cpu_outputs
);
int64_t
end
=
NowMicros
();
micros
=
end
-
start
;
}
...
...
@@ -240,7 +322,7 @@ int Main(int argc, char **argv) {
int64_t
micros
=
0
;
int64_t
start
=
NowMicros
();
for
(;
micros
<
run_micros
;
++
frames
)
{
gpu_engine
.
Run
(
input
_data
.
get
(),
input_shape_vec
,
gpu_output_data
.
get
()
);
gpu_engine
.
Run
(
input
s
,
&
gpu_outputs
);
int64_t
end
=
NowMicros
();
micros
=
end
-
start
;
}
...
...
@@ -254,7 +336,7 @@ int Main(int argc, char **argv) {
int64_t
micros
=
0
;
int64_t
start
=
NowMicros
();
for
(;
micros
<
run_micros
;
++
frames
)
{
dsp_engine
.
Run
(
input
_data
.
get
(),
input_shape_vec
,
dsp_output_data
.
get
()
);
dsp_engine
.
Run
(
input
s
,
&
dsp_outputs
);
int64_t
end
=
NowMicros
();
micros
=
end
-
start
;
}
...
...
tools/build_run_throughput_test.sh
浏览文件 @
1fdb5593
...
...
@@ -72,9 +72,11 @@ adb -s $DEVICE_ID </dev/null shell \
MACE_KERNEL_PATH
=
$KERNEL_DIR
\
MACE_LIMIT_OPENCL_KERNEL_TIME
=
${
LIMIT_OPENCL_KERNEL_TIME
}
\
${
PHONE_DATA_DIR
}
/model_throughput_test
\
--input_node
=
"
${
INPUT_NODES
}
"
\
--input_shape
=
"
${
INPUT_SHAPES
}
"
\
--output_node
=
"
${
OUTPUT_NODES
}
"
\
--output_shape
=
"
${
OUTPUT_SHAPES
}
"
\
--input_file
=
${
PHONE_DATA_DIR
}
/
${
INPUT_FILE_NAME
}
_
${
INPUT_NODES
}
\
--input_file
=
${
PHONE_DATA_DIR
}
/
${
INPUT_FILE_NAME
}
\
--cpu_model_data_file
=
${
PHONE_DATA_DIR
}
/
${
CPU_MODEL_TAG
}
.data
\
--gpu_model_data_file
=
${
PHONE_DATA_DIR
}
/
${
GPU_MODEL_TAG
}
.data
\
--dsp_model_data_file
=
${
PHONE_DATA_DIR
}
/
${
DSP_MODEL_TAG
}
.data
\
...
...
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