TConv3DBackpropInputTiling Structure

The TConv3DBackpropInputTiling structure contains the Conv3DBackpropInput operator specifications and parameters related to the tiling algorithm. It is passed to the Conv3DBackpropInput kernel for data tiling, data transfer, and computation. The following table describes the parameters in the TConv3DBackpropInputTiling structure.

You can call the GetTiling API to obtain the TConv3DBackpropInputTiling structure. For details, see Conv3DBackpropInput Tiling Usage Description. Currently, the parameters in the TConv3DBackpropInputTiling structure cannot be customized.

**Table 1** TConv3DBackpropInputTiling structure
Parameter	Description
batch	N of the GradOutput input, which is equal to the N of the forward convolution input.
cin	Channel of the GradInput output, which is equal to the channel of the forward convolution input.
cout	Channel of the GradOutput input.
cout1	C1 of the GradOutput input, which is equal to cout/c0.
cin1	C1 of the GradInput output, which is equal to the C1 of the forward convolution input, that is, cin/c0.
cout1G	Reserved parameter, which is imperceptible to users.
cin1G	Reserved parameter, which is imperceptible to users.
c0	Size of C0 in the current input data type. Currently, this parameter can only be set to 16.
c0Bits	Number of right shift bits equivalent to the number of bits divided by c0. For example, if c0 is 8, c0Bits is 3; if c0 is 16, c0Bits is 4.
dout	Depth of the input GradOutput. The unit is element.
ho	Height of the input GradOutput. The unit is element.
wo	Width of the input GradOutput. The unit is element.
di	Depth of the GradInput output, which is equal to the depth of the forward convolution input. The unit is element.
hi	Height size of the GradInput output, which is equal to the height size of the forward convolution input. The unit is element.
wi	Width size of the GradInput output, which is equal to the width size of the forward convolution input. The unit is element.
dk	Depth size of the input Weight. The unit is element.
hk	Height size of the input Weight. The unit is element.
wk	Width size of the input Weight. The unit is element.
group	Reserved parameter, which is imperceptible to users.
strideD	Depth size of the stride in backpropagation convolution computation. The unit is element.
strideH	Height size of the stride in backpropagation convolution computation. The unit is element.
strideW	Width size of the stride in backpropagation convolution computation. The unit is element.
padFront	Front direction of the depth dimension of the output matrix GradInput Padding in backpropagation convolution computation. The unit is element.
padBack	Backward direction of the depth dimension of the output matrix GradInput Padding in backpropagation convolution computation. The unit is element.
padUp	Upward direction of the height dimension of the output matrix GradInput Padding in backpropagation convolution computation. The unit is element.
padDown	Downward direction of the height dimension of the output matrix GradInput Padding in backpropagation convolution computation. The unit is element.
padLeft	Leftward direction of the width dimension of the output matrix GradInput Padding in backpropagation convolution computation. The unit is element.
padRight	Rightward direction of the width dimension of the output matrix GradInput Padding in backpropagation convolution computation. The unit is element.
backpropPadTail	Reserved parameter, which is imperceptible to users.
backpropPadUp	Upward direction of the height dimension of the input matrix GradOutput Padding in backpropagation convolution computation. The unit is element.
backpropPadDown	Downward direction of the height dimension of the input matrix GradOutput Padding in backpropagation convolution computation. The unit is element.
backpropPadLeft	Leftward direction of the width dimension of the input matrix GradOutput Padding in backpropagation convolution computation. The unit is element.
backpropPadRight	Rightward direction of the width dimension of the input matrix GradOutput Padding in backpropagation convolution computation. The unit is element.
dilationD	Depth size of the dilation in backpropagation convolution computation. The unit is element.
dilationH	Height size of the dilation in backpropagation convolution computation. The unit is element.
dilationW	Width size of the dilation in backpropagation convolution computation. The unit is element.
al0Pbuffer	The value 1 indicates that DoubleBuffer is disabled, and the value 2 indicates that DoubleBuffer is enabled.
bl0Pbuffer	The value 1 indicates that DoubleBuffer is disabled, and the value 2 indicates that DoubleBuffer is enabled.
cl0Pbuffer	The value 1 indicates that DoubleBuffer is disabled, and the value 2 indicates that DoubleBuffer is enabled.
al1Pbuffer	The value 1 indicates that DoubleBuffer is disabled, and the value 2 indicates that DoubleBuffer is enabled.
bl1Pbuffer	The value 1 indicates that DoubleBuffer is disabled, and the value 2 indicates that DoubleBuffer is enabled.
singleCoreGroup	Reserved parameter, which is imperceptible to users.
singleCoreCout	Size of the cout data to be computed in the M direction of a single core.
singleCoreCout1	Size of cout1 on a single core.
singleCoreCin1	Size of cin1 on a single core.
singleCoreDin	Size of Din on a single core.
singleCoreHo	Size of the ho data to be computed in the K direction of a single core.
baseM	Size in the M direction on L0.
baseK	Size in the K direction on L0.
baseN	Size in the N direction on L0.
baseD	Reserved parameter, which is imperceptible to users.
baseBatch	Reserved parameter, which is imperceptible to users.
baseGroup	Reserved parameter, which is imperceptible to users.
stepM	Multiple of baseM in the buffer M direction of the feature matrix cached in L1.
stepN	Multiple of baseN in the buffer N direction of the weight matrix cached in L1.
stepKa	Multiple of baseK in the buffer K direction of the feature matrix cached in L1.
stepKb	Multiple of baseK in the buffer K direction of the weight matrix cached in L1.
stepBatch	Reserved parameter, which is imperceptible to users.
stepGroup	Reserved parameter, which is imperceptible to users.
iterateOrder	Reserved parameter, which is imperceptible to users.
hf32Flag	Reserved parameter, which is imperceptible to users.
initOutputFlag	Reserved parameter, which is imperceptible to users.
reserved	Reserved parameter, which is imperceptible to users.
singleCoreBatch	Reserved parameter, which is imperceptible to users.
singleCoreM	Size of the data to be computed in the M direction of a single core.
singleCoreCin	Size of the cin data to be computed in the N direction of a single core.

Parent topic: Conv3DBackpropInput Tiling API