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
野马c
Mace
提交
2fef9af5
Mace
项目概览
野马c
/
Mace
与 Fork 源项目一致
Fork自
Xiaomi / Mace
通知
1
Star
1
Fork
0
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
2fef9af5
编写于
10月 26, 2017
作者:
L
liuqi
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Change the calculation of local work group size.
上级
845377f2
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
9 addition
and
10 deletion
+9
-10
mace/core/runtime/opencl/opencl_runtime.cc
mace/core/runtime/opencl/opencl_runtime.cc
+4
-4
mace/core/runtime/opencl/opencl_runtime.h
mace/core/runtime/opencl/opencl_runtime.h
+2
-2
mace/kernels/opencl/batch_norm_opencl.cc
mace/kernels/opencl/batch_norm_opencl.cc
+3
-4
未找到文件。
mace/core/runtime/opencl/opencl_runtime.cc
浏览文件 @
2fef9af5
...
@@ -160,16 +160,16 @@ cl::Program &OpenCLRuntime::program() {
...
@@ -160,16 +160,16 @@ cl::Program &OpenCLRuntime::program() {
return
program_
;
return
program_
;
}
}
in
t
OpenCLRuntime
::
GetDeviceMaxWorkGroupSize
()
{
uint32_
t
OpenCLRuntime
::
GetDeviceMaxWorkGroupSize
()
{
unsigned
long
long
size
=
0
;
unsigned
long
long
size
=
0
;
device_
.
getInfo
(
CL_DEVICE_MAX_WORK_GROUP_SIZE
,
&
size
);
device_
.
getInfo
(
CL_DEVICE_MAX_WORK_GROUP_SIZE
,
&
size
);
return
static_cast
<
in
t
>
(
size
);
return
static_cast
<
uint32_
t
>
(
size
);
}
}
in
t
OpenCLRuntime
::
GetKernelMaxWorkGroupSize
(
const
cl
::
Kernel
&
kernel
)
{
uint32_
t
OpenCLRuntime
::
GetKernelMaxWorkGroupSize
(
const
cl
::
Kernel
&
kernel
)
{
unsigned
long
long
size
=
0
;
unsigned
long
long
size
=
0
;
kernel
.
getWorkGroupInfo
(
device_
,
CL_KERNEL_WORK_GROUP_SIZE
,
&
size
);
kernel
.
getWorkGroupInfo
(
device_
,
CL_KERNEL_WORK_GROUP_SIZE
,
&
size
);
return
static_cast
<
in
t
>
(
size
);
return
static_cast
<
uint32_
t
>
(
size
);
}
}
}
// namespace mace
}
// namespace mace
mace/core/runtime/opencl/opencl_runtime.h
浏览文件 @
2fef9af5
...
@@ -21,8 +21,8 @@ class OpenCLRuntime {
...
@@ -21,8 +21,8 @@ class OpenCLRuntime {
public:
public:
static
OpenCLRuntime
*
Get
();
static
OpenCLRuntime
*
Get
();
in
t
GetDeviceMaxWorkGroupSize
();
uint32_
t
GetDeviceMaxWorkGroupSize
();
in
t
GetKernelMaxWorkGroupSize
(
const
cl
::
Kernel
&
kernel
);
uint32_
t
GetKernelMaxWorkGroupSize
(
const
cl
::
Kernel
&
kernel
);
cl
::
Context
&
context
();
cl
::
Context
&
context
();
cl
::
Device
&
device
();
cl
::
Device
&
device
();
cl
::
CommandQueue
&
command_queue
();
cl
::
CommandQueue
&
command_queue
();
...
...
mace/kernels/opencl/batch_norm_opencl.cc
浏览文件 @
2fef9af5
...
@@ -22,13 +22,15 @@ void BatchNormFunctor<DeviceType::OPENCL, float>::operator()(
...
@@ -22,13 +22,15 @@ void BatchNormFunctor<DeviceType::OPENCL, float>::operator()(
const
uint32_t
gws
[
3
]
=
{
static_cast
<
uint32_t
>
(
input
->
dim
(
0
)),
const
uint32_t
gws
[
3
]
=
{
static_cast
<
uint32_t
>
(
input
->
dim
(
0
)),
static_cast
<
uint32_t
>
(
input
->
dim
(
1
)),
static_cast
<
uint32_t
>
(
input
->
dim
(
1
)),
static_cast
<
uint32_t
>
(
input
->
dim
(
2
)
*
input
->
dim
(
3
))};
static_cast
<
uint32_t
>
(
input
->
dim
(
2
)
*
input
->
dim
(
3
))};
const
uint32_t
lws
[
3
]
=
{
1
,
2
,
128
};
auto
runtime
=
OpenCLRuntime
::
Get
();
auto
runtime
=
OpenCLRuntime
::
Get
();
auto
program
=
runtime
->
program
();
auto
program
=
runtime
->
program
();
auto
bm_kernel
=
cl
::
Kernel
(
program
,
"batch_norm"
);
auto
bm_kernel
=
cl
::
Kernel
(
program
,
"batch_norm"
);
const
uint32_t
kwg_size
=
runtime
->
GetKernelMaxWorkGroupSize
(
bm_kernel
);
const
uint32_t
lws
[
3
]
=
{
1
,
kwg_size
/
128
,
128
};
uint32_t
idx
=
0
;
uint32_t
idx
=
0
;
bm_kernel
.
setArg
(
idx
++
,
*
(
static_cast
<
const
cl
::
Buffer
*>
(
input
->
buffer
())));
bm_kernel
.
setArg
(
idx
++
,
*
(
static_cast
<
const
cl
::
Buffer
*>
(
input
->
buffer
())));
bm_kernel
.
setArg
(
idx
++
,
*
(
static_cast
<
cl
::
Buffer
*>
(
scale
->
buffer
())));
bm_kernel
.
setArg
(
idx
++
,
*
(
static_cast
<
cl
::
Buffer
*>
(
scale
->
buffer
())));
...
@@ -41,9 +43,6 @@ void BatchNormFunctor<DeviceType::OPENCL, float>::operator()(
...
@@ -41,9 +43,6 @@ void BatchNormFunctor<DeviceType::OPENCL, float>::operator()(
bm_kernel
.
setArg
(
idx
++
,
lws
[
1
]
*
sizeof
(
float
),
nullptr
);
bm_kernel
.
setArg
(
idx
++
,
lws
[
1
]
*
sizeof
(
float
),
nullptr
);
bm_kernel
.
setArg
(
idx
++
,
lws
[
1
]
*
sizeof
(
float
),
nullptr
);
bm_kernel
.
setArg
(
idx
++
,
lws
[
1
]
*
sizeof
(
float
),
nullptr
);
MACE_CHECK
(
std
::
accumulate
(
lws
,
lws
+
3
,
1
,
std
::
multiplies
<
uint32_t
>
())
<
runtime
->
GetKernelMaxWorkGroupSize
(
bm_kernel
));
cl_int
error
=
runtime
->
command_queue
().
enqueueNDRangeKernel
(
cl_int
error
=
runtime
->
command_queue
().
enqueueNDRangeKernel
(
bm_kernel
,
cl
::
NullRange
,
bm_kernel
,
cl
::
NullRange
,
cl
::
NDRange
(
gws
[
0
],
gws
[
1
],
gws
[
2
]),
cl
::
NDRange
(
gws
[
0
],
gws
[
1
],
gws
[
2
]),
...
...
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