Optimize conv1x1
- Tensor's Clear only clears its raw_size instead of buffer size - Don't pad conv1x1 - Rearrange conv1x1's assembly codes and optimize loading strategy
Showing
想要评论请 注册 或 登录
Fork自 Xiaomi / Mace
- Tensor's Clear only clears its raw_size instead of buffer size - Don't pad conv1x1 - Rearrange conv1x1's assembly codes and optimize loading strategy