Skip to content
This repository has been archived by the owner on Jan 24, 2024. It is now read-only.

How to use api of direct_conv_xxx in sass_func.hpp #525

Open
peyer opened this issue Jun 17, 2019 · 2 comments
Open

How to use api of direct_conv_xxx in sass_func.hpp #525

peyer opened this issue Jun 17, 2019 · 2 comments

Comments

@peyer
Copy link

peyer commented Jun 17, 2019

For convenience, I just take an instance below.
If the shape of input tensor of conv op is [1, 32, 110, 94] ([N, C, H, W] order),
the shape of kernel of conv op is 33([H, W] order), the channels of kernel is 64, the stride of kernel is 11([H, W] order),
the pad of kernel is 00, the dilation of kernel is 11, the group of conv op is 1,
then how to set parameters below
img_in_channel_stride
img_in_height_stride
img_in_width_stride
img_out_channel_stride
img_out_height_stride
img_out_width_stride

@xyoungli
Copy link
Collaborator

For convenience, I just take an instance below.
If the shape of input tensor of conv op is [1, 32, 110, 94] ([N, C, H, W] order),
the shape of kernel of conv op is 3_3([H, W] order), the channels of kernel is 64, the stride of kernel is 1_1([H, W] order),
the pad of kernel is 0_0, the dilation of kernel is 1_1, the group of conv op is 1,
then how to set parameters below
img_in_channel_stride
img_in_height_stride
img_in_width_stride
img_out_channel_stride
img_out_height_stride
img_out_width_stride

if tensor is continous, then:
channel_stride = w * h
height_stride = w
width_stride = 1

@peyer
Copy link
Author

peyer commented Jun 18, 2019

@xyoungli I changed the value of parameter according your advice, but I still got different result compared with cudnn, could you take some time to check my calling code, I have sent it to you on email.Thanks a lot!

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants