High Performance Implementation of 3D Convolutional Neural Networks on a GPU

Algorithm 2

Convolutional layer implemented with WMFA (, ).

is the number of image tiles.

is the input tile size.

Neighbouring tiles overlap by .

is input tile in channel .

is filter in channel .

is output tile in filter .

for   to   do

for   to C  do

Scatter to matrices :

end for

end for

for   to   do

for   to C  do

Scatter to matrices :

end for

end for

for   to   do

for   to   do

for   to   do

end for

end for

end for

for   to   do

for   to   do

Gather from matrices

end for

end for