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
booniebears
Paddle
提交
613303db
P
Paddle
项目概览
booniebears
/
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 >>
未验证
提交
613303db
编写于
6月 10, 2020
作者:
W
wangchaochaohu
提交者:
GitHub
6月 10, 2020
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
refine the slice Op to improve the performance of xlnet for fp16 training (#24967)
上级
37bdb526
变更
1
隐藏空白更改
内联
并排
Showing
1 changed file
with
17 addition
and
12 deletion
+17
-12
paddle/fluid/operators/slice_op.cu
paddle/fluid/operators/slice_op.cu
+17
-12
未找到文件。
paddle/fluid/operators/slice_op.cu
浏览文件 @
613303db
...
...
@@ -13,12 +13,12 @@ See the License for the specific language governing permissions and
limitations under the License. */
#include <thrust/device_vector.h>
#include "paddle/fluid/framework/tensor_util.h"
#include "paddle/fluid/operators/math/math_function.h"
#include "paddle/fluid/operators/slice_op.h"
#include "paddle/fluid/platform/cuda_device_function.h"
#include "paddle/fluid/platform/cuda_primitives.h"
#include "paddle/fluid/platform/float16.h"
namespace
paddle
{
namespace
operators
{
...
...
@@ -94,17 +94,22 @@ class SliceGradKernel<paddle::platform::CUDADeviceContext,
dim3
blocks
((
numel
-
1
)
/
PADDLE_CUDA_NUM_THREADS
+
1
);
dim3
threads
(
PADDLE_CUDA_NUM_THREADS
);
auto
stream
=
ctx
.
cuda_device_context
().
stream
();
auto
out_shape
=
framework
::
vectorize
<
int64_t
>
(
out_dims
);
thrust
::
device_vector
<
int64_t
>
out_dims_vec
(
out_shape
.
begin
(),
out_shape
.
end
());
auto
in_shape
=
framework
::
vectorize
<
int64_t
>
(
in_dims
);
thrust
::
device_vector
<
int64_t
>
in_dims_vec
(
in_shape
.
begin
(),
in_shape
.
end
());
thrust
::
device_vector
<
int64_t
>
offsets_vec
(
offsets
.
begin
(),
offsets
.
end
());
const
int64_t
*
out_dims_ptr
=
thrust
::
raw_pointer_cast
(
out_dims_vec
.
data
());
const
int64_t
*
in_dims_ptr
=
thrust
::
raw_pointer_cast
(
in_dims_vec
.
data
());
const
int64_t
*
offsets_ptr
=
thrust
::
raw_pointer_cast
(
offsets_vec
.
data
());
const
std
::
vector
<
int64_t
>
out_shape
=
framework
::
vectorize
<
int64_t
>
(
out_dims
);
const
std
::
vector
<
int64_t
>
in_shape
=
framework
::
vectorize
<
int64_t
>
(
in_dims
);
framework
::
Tensor
out_dims_tensor
;
framework
::
Tensor
in_dims_tensor
;
framework
::
Tensor
offsets_tensor
;
framework
::
TensorFromVector
(
out_shape
,
ctx
.
device_context
(),
&
out_dims_tensor
);
framework
::
TensorFromVector
(
in_shape
,
ctx
.
device_context
(),
&
in_dims_tensor
);
framework
::
TensorFromVector
(
offsets
,
ctx
.
device_context
(),
&
offsets_tensor
);
const
int64_t
*
out_dims_ptr
=
out_dims_tensor
.
data
<
int64_t
>
();
const
int64_t
*
in_dims_ptr
=
in_dims_tensor
.
data
<
int64_t
>
();
const
int64_t
*
offsets_ptr
=
offsets_tensor
.
data
<
int64_t
>
();
switch
(
rank
)
{
case
1
:
...
...
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