inferenc cost time (2048x600x3) < (2048x512x3)

[w1005444804]

error log | 日志或报错信息 | ログ

context | 编译/运行环境 | バックグラウンド

win32 ncnn build myself, create my net using interface create_layer("") one by one.

how to reproduce | 复现步骤 | 再現方法

more | 其他 | その他

[DC-Zhou]

OK,Sir make sure what operation your net use; in sonme hardware operation like convolution, ncnn maybe use pack,which means Change memory layout to accommodate L1 cache of hardware,or other Acceleration instruction set such as AVX or SSE.so if you want to check time consuming, check your net each operation

[w1005444804]

OK,Sir make sure what operation your net use; in sonme hardware operation like convolution, ncnn maybe use pack,which means Change memory layout to accommodate L1 cache of hardware,or other Acceleration instruction set such as AVX or SSE.so if you want to check time consuming, check your net each operation

win32， i7, infer on cpu， Finally found that depthwise convolution is very slow，it is slower than standard convolution，it may lead to the final time-consuming and the image input area is not proportional