PadV4Grad

English | 简体中文

Product Support Status

Product	Supported
Ascend 950PR&950DT products	×
Atlas A3 products	√
Atlas A2 products	√
Atlas 200I/500 A2 inference products	×
Atlas inference products	×
Atlas training products	×

Function Description

• Operator function: backpropagation of reflection_pad1d/reflection_pad2d.

Parameter Description

Parameter Name	Input/Output/Attribute	Description	Data Type	Data Format
gradOutput	Input	Forward propagation of reflection_pad1d/reflection_pad2d. The shape supports 2 to 4 dimensions, and the dimensions must be consistent with self and gradInput. The backward operator of reflection_pad1d corresponds to 2/3 dimensions, and the backward operator of reflection_pad2d corresponds to 3/4 dimensions.	BFLOAT16, FLOAT16, FLOAT32, DOUBLE, COMPLEX64, COMPLEX128	ND
self	Input	The shape supports 2 to 4 dimensions, and the dimensions must be consistent with self and gradInput. The shape is the same as that of gradInput.	BFLOAT16, FLOAT16, FLOAT32, DOUBLE, COMPLEX64, COMPLEX128	ND
padding	Input	padding describes the size of outward padding, and its length is 2 or 4. When the length is 2, it indicates the values to be padded on the left and right for 1d, and the first two values of padding must be smaller than the value of the last dimension of self. When the length is 4, it indicates the values to be padded on the left, right, top, and bottom for 2d. The first two values of padding must be smaller than the value of the last dimension of self, and the last two values must be smaller than the value of the second-to-last dimension.	INT64
gradInput	Output	The data type is the same as that of self.	BFLOAT16, FLOAT16, FLOAT32, DOUBLE, COMPLEX64, COMPLEX128	ND

Constraint Description

• Input shape restriction: the dimensions of gradOutput, self, and gradInput must be consistent (three or four dimensions are supported), and their shapes must be consistent with the output shape of the forward propagation of reflection_pad1d/reflection_pad2d.

• Input value range restriction: when the length of padding is 2, the first two values of padding must be smaller than the value of the last dimension of self; when the length is 4, the first two values of padding must be smaller than the value of the last dimension of self, and the last two values must be smaller than the value of the second-to-last dimension.

Invocation Description

Invocation Method	Invocation Sample	Description
aclnn invocation	test_aclnn_reflection_pad1d_backward.cpp	Invokes the PadV4Grad operator through the aclnnReflectionPad1dBackward API.