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
吃玉米的猫
Paddle
提交
e5ce9659
P
Paddle
项目概览
吃玉米的猫
/
Paddle
与 Fork 源项目一致
Fork自
PaddlePaddle / Paddle
通知
1
Star
0
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
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
0
Issue
0
列表
看板
标记
里程碑
合并请求
0
合并请求
0
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 >>
提交
e5ce9659
编写于
10月 19, 2018
作者:
T
tensor-tang
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
refine and add eltadd_relu unit test
上级
7cb19a59
变更
3
隐藏空白更改
内联
并排
Showing
3 changed file
with
58 addition
and
4 deletion
+58
-4
paddle/fluid/operators/math/fc_compute.h
paddle/fluid/operators/math/fc_compute.h
+1
-1
paddle/fluid/operators/math/jit_kernel_blas.cc
paddle/fluid/operators/math/jit_kernel_blas.cc
+0
-3
paddle/fluid/operators/math/jit_kernel_test.cc
paddle/fluid/operators/math/jit_kernel_test.cc
+57
-0
未找到文件。
paddle/fluid/operators/math/fc_compute.h
浏览文件 @
e5ce9659
...
...
@@ -15,7 +15,7 @@ limitations under the License. */
#pragma once
#include "paddle/fluid/operators/math/blas.h"
#include "paddle/fluid/operators/math/jit_kernel.h"
// TODO(TJ): add deps
#include "paddle/fluid/operators/math/jit_kernel.h"
DECLARE_int32
(
paddle_num_threads
);
...
...
paddle/fluid/operators/math/jit_kernel_blas.cc
浏览文件 @
e5ce9659
...
...
@@ -447,20 +447,17 @@ class VAddReluKernelImpl : public VAddReluKernel<T> {
#ifdef __AVX__
INTRI8_FLOAT
(
jit
::
avx
);
INTRI16_FLOAT
(
jit
::
avx
);
INTRI_COMMON_FLOAT
(
jit
::
avx
,
kGT8LT16
);
INTRI_COMMON_FLOAT
(
jit
::
avx
,
kGT16
);
#endif
#ifdef __AVX2__
INTRI8_FLOAT
(
jit
::
avx2
);
INTRI16_FLOAT
(
jit
::
avx2
);
INTRI_COMMON_FLOAT
(
jit
::
avx2
,
kGT8LT16
);
INTRI_COMMON_FLOAT
(
jit
::
avx2
,
kGT16
);
#endif
#ifdef __AVX512F__
// TODO(TJ): refine avx512
INTRI8_FLOAT
(
jit
::
avx512f
);
INTRI16_FLOAT
(
jit
::
avx512f
);
INTRI_COMMON_FLOAT
(
jit
::
avx512f
,
kGT8LT16
);
INTRI_COMMON_FLOAT
(
jit
::
avx512f
,
kGT16
);
#endif
...
...
paddle/fluid/operators/math/jit_kernel_test.cc
浏览文件 @
e5ce9659
...
...
@@ -712,6 +712,63 @@ TEST(JitKernel, vadd) {
}
}
void
vaddrelu_ref
(
const
int
n
,
const
float
*
x
,
const
float
*
y
,
float
*
z
)
{
for
(
int
i
=
0
;
i
<
n
;
++
i
)
{
z
[
i
]
=
x
[
i
]
+
y
[
i
];
z
[
i
]
=
z
[
i
]
>
0
?
z
[
i
]
:
0
;
}
}
void
vaddrelu_better
(
const
std
::
shared_ptr
<
const
paddle
::
operators
::
math
::
jitkernel
::
VAddKernel
<
float
>>&
vadd
,
const
std
::
shared_ptr
<
const
paddle
::
operators
::
math
::
jitkernel
::
VReluKernel
<
float
>>&
vrelu
,
const
float
*
x
,
const
float
*
y
,
float
*
z
)
{
vadd
->
Compute
(
x
,
y
,
z
);
vrelu
->
Compute
(
z
,
z
);
}
TEST
(
JitKernel
,
vaddrelu
)
{
namespace
jit
=
paddle
::
operators
::
math
::
jitkernel
;
for
(
int
d
:
{
7
,
8
,
15
,
16
,
30
,
256
,
512
})
{
std
::
vector
<
float
>
x
(
d
),
y
(
d
);
std
::
vector
<
float
>
zref
(
d
),
ztgt
(
d
);
RandomVec
<
float
>
(
d
,
x
.
data
());
RandomVec
<
float
>
(
d
,
y
.
data
());
const
auto
&
ker
=
jit
::
KernelPool
::
Instance
().
template
Get
<
jit
::
VAddReluKernel
<
float
>
>
(
d
);
const
auto
&
vadd
=
jit
::
KernelPool
::
Instance
().
template
Get
<
jit
::
VAddKernel
<
float
>
>
(
d
);
const
auto
&
vrelu
=
jit
::
KernelPool
::
Instance
().
template
Get
<
jit
::
VReluKernel
<
float
>
>
(
d
);
const
float
*
x_data
=
x
.
data
();
const
float
*
y_data
=
y
.
data
();
float
*
ztgt_data
=
ztgt
.
data
();
float
*
zref_data
=
zref
.
data
();
auto
trefs
=
GetCurrentUS
();
for
(
int
i
=
0
;
i
<
repeat
;
++
i
)
{
vadd_ref
(
d
,
x_data
,
y_data
,
zref_data
);
}
auto
trefe
=
GetCurrentUS
();
auto
tmkls
=
GetCurrentUS
();
for
(
int
i
=
0
;
i
<
repeat
;
++
i
)
{
vaddrelu_better
(
vadd
,
vrelu
,
x_data
,
y_data
,
zref_data
);
}
auto
tmkle
=
GetCurrentUS
();
auto
ttgts
=
GetCurrentUS
();
for
(
int
i
=
0
;
i
<
repeat
;
++
i
)
{
ker
->
Compute
(
x_data
,
y_data
,
ztgt_data
);
}
auto
ttgte
=
GetCurrentUS
();
VLOG
(
3
)
<<
"Vec size "
<<
d
<<
": refer takes: "
<<
(
trefe
-
trefs
)
/
repeat
<<
" us, better takes: "
<<
(
tmkle
-
tmkls
)
/
repeat
<<
" us, "
<<
"tgt takes: "
<<
(
ttgte
-
ttgts
)
/
repeat
;
for
(
int
i
=
0
;
i
<
d
;
++
i
)
{
EXPECT_NEAR
(
ztgt_data
[
i
],
zref_data
[
i
],
1e-3
);
}
}
}
TEST
(
JitKernel
,
pool
)
{
namespace
jit
=
paddle
::
operators
::
math
::
jitkernel
;
const
int
frame_size
=
4
;
...
...
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