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
TonyTonyFun
Paddle
提交
bce72c7f
P
Paddle
项目概览
TonyTonyFun
/
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
bce72c7f
编写于
8月 12, 2019
作者:
J
joanna.wozna.intel
提交者:
Tao Luo
8月 12, 2019
浏览文件
操作
浏览文件
下载
电子邮件补丁
差异文件
Replace Relu with bounded Relu in MobileNetV2 quantization (#18988)
test=develop
上级
e044e842
变更
2
隐藏空白更改
内联
并排
Showing
2 changed file
with
13 addition
and
9 deletion
+13
-9
paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc
paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc
+9
-0
paddle/fluid/operators/mkldnn/conv_mkldnn_op.cc
paddle/fluid/operators/mkldnn/conv_mkldnn_op.cc
+4
-9
未找到文件。
paddle/fluid/framework/ir/mkldnn/cpu_quantize_pass.cc
浏览文件 @
bce72c7f
...
...
@@ -208,6 +208,15 @@ void CPUQuantizePass::QuantizeConv(Graph* graph,
DequantizeOutput
(
g
,
conv_op
,
conv_output
,
"Output"
,
output_scale
,
is_output_unsigned
,
"Scale_out"
);
// change threshold in bounded ReLu
if
(
conv_op
->
Op
()
->
HasAttr
(
"fuse_brelu"
)
&&
boost
::
get
<
bool
>
(
conv_op
->
Op
()
->
GetAttr
(
"fuse_brelu"
)))
{
float
scale_out
=
boost
::
get
<
float
>
(
conv_op
->
Op
()
->
GetAttr
(
"Scale_out"
));
float
threshold
=
boost
::
get
<
float
>
(
conv_op
->
Op
()
->
GetAttr
(
"fuse_brelu_threshold"
));
conv_op
->
Op
()
->
SetAttr
(
"fuse_brelu_threshold"
,
scale_out
*
threshold
);
}
++
quantize_conv_count
;
};
...
...
paddle/fluid/operators/mkldnn/conv_mkldnn_op.cc
浏览文件 @
bce72c7f
...
...
@@ -484,9 +484,6 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
handler
.
reset
(
new
platform
::
ConvMKLDNNHandler
(
dev_ctx
,
mkldnn_engine
,
key
));
// create a conv primitive descriptor and save it for usage in backward
// TODO(lidanqing): We use relu post-op instead of brelu post-op cause
// mkldnn v0.18 does not support INT8 brelu post-op. Use code in /**/ when
// v0.20 is enabled
auto
propagation
=
is_test
?
mkldnn
::
prop_kind
::
forward_scoring
:
mkldnn
::
prop_kind
::
forward_training
;
...
...
@@ -496,15 +493,13 @@ class ConvMKLDNNOpKernel : public paddle::framework::OpKernel<T> {
mkldnn
::
memory
::
format
::
x
);
conv_pd
=
handler
->
AcquireConvolutionPrimitiveDescriptor
(
src_md
,
weights_md
,
bias_md
,
dst_md
,
strides
,
paddings
,
mkldnn_engine
,
fuse_relu
||
fuse_brelu
/*fuse_relu*/
,
fuse_residual_conn
,
false
/*fuse_brelu*/
,
fuse_brelu_threshold
,
propagation
,
output_shift_scale
,
sum_scale
);
mkldnn_engine
,
fuse_relu
,
fuse_residual_conn
,
fuse_brelu
,
fuse_brelu_threshold
,
propagation
,
output_shift_scale
,
sum_scale
);
}
else
{
conv_pd
=
handler
->
AcquireConvolutionPrimitiveDescriptor
(
src_md
,
weights_md
,
boost
::
none
,
dst_md
,
strides
,
paddings
,
mkldnn_engine
,
fuse_relu
||
fuse_brelu
/*fuse_relu*/
,
fuse_residual_conn
,
false
/*fuse_brelu*/
,
fuse_brelu_threshold
,
propagation
,
output_shift_scale
,
sum_scale
);
mkldnn_engine
,
fuse_relu
,
fuse_residual_conn
,
fuse_brelu
,
fuse_brelu_threshold
,
propagation
,
output_shift_scale
,
sum_scale
);
}
// create mkldnn memory from input tensors (data/weights)
...
...
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