Mojo struct
ConvShape
@register_passable(trivial)
struct ConvShape[rank: Int]
A shape struct describing the convolution dimensions.
Fields
- n (
Int): - input_dims (
Index[rank]): - output_dims (
Index[rank]): - filter_dims (
Index[rank]): - c (
Int): - f (
Int): - stride (
Index[rank]): - dilation (
Index[rank]): - pad_d (
Index[2]): - pad_h (
Index[2]): - pad_w (
Index[2]): - num_groups (
Int):
Implemented traits
AnyType,
Copyable,
ExplicitlyCopyable,
Movable,
UnknownDestructibility
Methods
d
d(self) -> Int
Input depth.
h
h(self) -> Int
Input height.
w
w(self) -> Int
Input width.
do
do(self) -> Int
Output depth.
ho
ho(self) -> Int
Output height.
wo
wo(self) -> Int
Output width.
q
q(self) -> Int
Filter window depth.
r
r(self) -> Int
Filter window height.
s
s(self) -> Int
Filter windown width.
filter_window_flat_size
filter_window_flat_size(self) -> Int
input_image_flat_size
input_image_flat_size(self) -> Int
output_image_flat_size
output_image_flat_size(self) -> Int
output_space_dims
output_space_dims(self) -> Index[rank]
output_flat_coord_to_input_offset
output_flat_coord_to_input_offset(self, n: Int, output_flat_coord: Int) -> Int
matmul_M
matmul_M(self) -> Int
matmul_N
matmul_N(self) -> Int
matmul_K
matmul_K(self) -> Int
padded
padded(self) -> Bool
c_per_group
c_per_group(self) -> Int
Returns the number of channels per group. Channel count must be divisible by group size.
f_per_group
f_per_group(self) -> Int
Returns the number of filters per group. Filter count must be divisible by group size.
f_to_group
f_to_group(self, f_idx: Int) -> Int
Given a global filter idx, returns the group idx of the group the filter belongs to.
c_to_group
c_to_group(self, c_idx: Int) -> Int
Given a global channel idx, returns the group idx of the group the channel belongs to.
f_in_group
f_in_group(self, f_idx: Int) -> Int
Given a global filter idx, returns the offset of the filter in its group.
c_in_group
c_in_group(self, c_idx: Int) -> Int
Given a global channel idx, returns the offset of the channel in its group.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!