未验证 提交 9e38adc8 编写于 作者: H HappyAngel 提交者: GitHub

[arm] improve con3x3_dw (#4063)


* optimize conv_dw profiler

* fix build conv_dw_3x3s1 bug

* update conv_dw_3x3s2

* fxi foormat test=develop
上级 ee4cb1dc
......@@ -323,6 +323,118 @@ void conv_depthwise_3x3s2p1_bias_s_relu(float* dout,
const int w_out,
ARMContext* ctx);
void conv_depthwise_3x3s1p0_bias_no_relu(float* dout,
const float* din,
const float* weights,
const float* bias,
bool flag_bias,
bool flag_relu,
const int num,
const int ch_in,
const int h_in,
const int w_in,
const int h_out,
const int w_out,
ARMContext* ctx);
void conv_depthwise_3x3s1p0_bias_s_no_relu(float* dout,
const float* din,
const float* weights,
const float* bias,
bool flag_bias,
bool flag_relu,
const int num,
const int ch_in,
const int h_in,
const int w_in,
const int h_out,
const int w_out,
ARMContext* ctx);
void conv_depthwise_3x3s1p1_bias_no_relu(float* dout,
const float* din,
const float* weights,
const float* bias,
bool flag_bias,
bool flag_relu,
const int num,
const int ch_in,
const int h_in,
const int w_in,
const int h_out,
const int w_out,
ARMContext* ctx);
void conv_depthwise_3x3s1p1_bias_s_no_relu(float* dout,
const float* din,
const float* weights,
const float* bias,
bool flag_bias,
bool flag_relu,
const int num,
const int ch_in,
const int h_in,
const int w_in,
const int h_out,
const int w_out,
ARMContext* ctx);
void conv_depthwise_3x3s2p0_bias_no_relu(float* dout,
const float* din,
const float* weights,
const float* bias,
bool flag_bias,
bool flag_relu,
const int num,
const int ch_in,
const int h_in,
const int w_in,
const int h_out,
const int w_out,
ARMContext* ctx);
void conv_depthwise_3x3s2p0_bias_s_no_relu(float* dout,
const float* din,
const float* weights,
const float* bias,
bool flag_bias,
bool flag_relu,
const int num,
const int ch_in,
const int h_in,
const int w_in,
const int h_out,
const int w_out,
ARMContext* ctx);
void conv_depthwise_3x3s2p1_bias_no_relu(float* dout,
const float* din,
const float* weights,
const float* bias,
bool flag_bias,
bool flag_relu,
const int num,
const int ch_in,
const int h_in,
const int w_in,
const int h_out,
const int w_out,
ARMContext* ctx);
void conv_depthwise_3x3s2p1_bias_s_no_relu(float* dout,
const float* din,
const float* weights,
const float* bias,
bool flag_bias,
bool flag_relu,
const int num,
const int ch_in,
const int h_in,
const int w_in,
const int h_out,
const int w_out,
ARMContext* ctx);
} // namespace math
} // namespace arm
} // namespace lite
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册