Is it a bug in the reduction layer?

12 views

Skip to first unread message

Chen Feng

unread,

Jun 20, 2017, 9:09:54 PM6/20/17

to Caffe Users, Teng-Yok Lee

When I use the reduction layer, it seems much slower than other layers on GPU mode. Checking the CUDA implementation, I found this line a little bit weird:

https://github.com/intel/caffe/blob/master/src/caffe/layers/reduction_layer.cu#L16

Why mutable_cpu_data, instead of mutable_gpu_data? Anyone know the reason? Thanks in advance!

Reply all

Reply to author

Forward

0 new messages