ConvolutionBackwardData#

ConvolutionBackwardData operation accepts $diff_dst$ , weights and optional dst shape as inputs, and compute the $diff_src$ .

If auto_pad attribute is specified to one of valid, same_upper and same_lower, pads_begin and pads_end attributes will be ignored. The paddings will be calculated by following the below formula:

Let the parameters be:

Parameter	Depth	Height	Width	Comment
Paddings: Front, top, and left	$P D_{L}$	$P H_{L}$	$P W_{L}$	In the attributes we use `pads_begin` to indicate the corresponding vector of paddings
Padding: Back, bottom, and right	$P D_{R}$	$P H_{R}$	$P W_{R}$	In the attributes we use `pads_end` to indicate the corresponding vector of paddings
Stride	$S D$	$S H$	$S W$	In the attributes we use `strides` to indicate the corresponding vector of strides
Dilation	$D D$	$D H$	$D W$	In the attributes we use `dilations` to indicate the corresponding vector of dilations

Firstly, $t o t a l_p a d d i n g$ is calculated according to $s r c_s h a p e$ and $d s t_s h a p e$ . Let $s r c_h$ be height dimension of $s r c_s h a p e$ and $d s t_h$ be height dimension of $d s t_s h a p e$ .

t o t a l_p a d d i n g_{h} = S H \times (s r c_{h} - 1) + ((K H - 1) \times D H + 1) - {dst}_{h} + o u t p u t_p a d d i n g_{h}

If auto_pad attribute is specified as valid:

\begin{array}{r} P D_{L} = 0 \\ P D_{R} = 0 \end{array}

If auto_pad attribute is specified as same_lower:

\begin{array}{r} P D_{L} = f l o o r (t o t a l_p a d d i n g / 2) \\ P D_{R} = t o t a l_p a d d i n g - P D_{L} \end{array}

If auto_pad attribute is specified as same_upper:

\begin{array}{r} P D_{L} = t o t a l_p a d d i n g - P D_{R} \\ P D_{R} = f l o o r (t o t a l_p a d d i n g / 2) \end{array}

where:

$d s t_s h a p e$ is either an attribute or an input tensor,
$o u t p u t_p a d d i n g$ is an optional attribute.

Operation Attributes#

Attribute Name	Description	Value Type	Supported Values	Required or Optional
`strides`	Controls the strides the weights tensor is moved when computing convolution	s64	A s64 list containing positive values	Required
`pads_begin`	Controls number of zeros to be add to the front/top/left of spatial dimensions, the attribute will be ignored when `auto_pad` attribute is specified to `same_upper`, `same_lower` or `valid`	s64	A s64 list containing non-negative values	Required
`pads_end`	Controls number of zeros to be add to the back/bottom/right of spatial dimensions, the attribute will be ignored when `auto_pad` attribute is specified to `same_upper`, `same_lower` or `valid`	s64	A s64 list containing non-negative values	Required
`dilations`	Controls the amount of stretching the kernel before convolution	s64	A s64 list containing positive values (>1 means dilated convolution)	Required
`auto_pad`	Controls how the padding is calculated	string	`none` (default), `same_upper`, `same_lower`, `valid`	Optional
`output_padding`	Adds additional amount of padding per each spatial axis in `dst`	s64	A s64 list containing non-negative values, all zeros by default	Optional
`groups`	Controls how input channels and output channels are divided into	s64	A positive s64 value, `1` by default	Optional
`data_format`	Controls how to interpret the shape of `src` and `dst`.	string	`NCX`, `NXC` (default)	Optional
`weights_format`	Controls how to interpret the shape of `weights`	string	`OIX`, `XIO` (default)	Optional
`dst_shape`	Denotes the shape of the `dst` tensor	s64	A s64 list containing positive values	Optional

Execution Arguments#

The inputs and outputs must be provided according to the below index order when constructing an operation.

Inputs#

Index	Argument Name	Required or Optional
0	`diff_dst`	Required
1	`weights`	Required
2	`dst_shape`	Optional

@note The shape of weights is $(o u t_c h a n n e l s, i n_c h a n n e l s / g r o u p s, s p a t i a l_s h a p e)$ for OIX format or $(s p a t i a l_s h a p e, i n_c h a n n e l s / g r o u p s, o u t_c h a n n e l s)$ for XIO format. Both $i n_c h a n n e l s$ and $o u t_c h a n n e l s$ must be divisible by groups attribute.

@note Either dst_shape input or dst_shape attribute should be provided. If both provided, dst_shape input will precede over dst_shape attribute.

Outputs#

Index	Argument Name	Required or Optional
0	`diff_src`	Required

Supported Data Types#

ConvolutionBackwardData operation supports the following data type combinations.

Diff_dst	Weights	Diff_src	Dst_shape
f32	f32	f32	s32
bf16	bf16	bf16	s32
f16	f16	f16	s32

ConvolutionBackwardData

Contents