提交 d0063b37 编写于 作者: xiebaiyuan's avatar xiebaiyuan

use gemm to s1p0 in conv add bn relu

上级 1743d3f3
......@@ -120,7 +120,8 @@ void ConvAddBNReluCompute(const FusionConvAddBNReluParam<CPU> &param) {
if (param.Groups() == param.Input()->dims()[1] &&
param.Input()->dims()[1] == param.Output()->dims()[1] &&
param.Filter()->dims()[2] == param.Filter()->dims()[3] &&
param.Filter()->dims()[2] == 3 && param.Strides()[0] == 1) {
param.Filter()->dims()[2] == 3 && param.Strides()[0] == 1 &&
param.paddings_[0] == 1) {
math::DepthwiseConvAddBNRelu3x3s1p1(param.Input(), param.Filter(),
param.Output(), param.NewScale(),
param.NewBias(), true);
......
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