Cdist

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: computes the p-norm distance between each pair of vectors in two sets of vectors.

• Computation formula: For the input tensors x1 and x2, their shapes are $[B,P,M]$ and $[B,R,M]$ respectively, and the shape of the output tensor y is $[B,P,R]$.

  When $p>0$:

  $$y_{b,i,j}={\left(\sum _{k=1}^M|x{1}_{b,i,k}-x{2}_{b,j,k}{|}^p\right)}^{\frac{1}{p}}$$

  When $p=0$, the output is the number of unequal elements in the corresponding vectors; when $p=+\infty$, the output is the maximum of the absolute differences between the elements of the corresponding vectors.

Parameter Description

Parameter Name	Input/Output/Attribute	Description	Data Type	Data Format
x1	Input	The first input tensor, whose shape is [B, P, M].	BFLOAT16, FLOAT16, FLOAT	ND
x2	Input	The second input tensor, whose shape is [B, R, M].	BFLOAT16, FLOAT16, FLOAT	ND
y	Output	p-norm distance computation result, whose shape is [B, P, R].	BFLOAT16, FLOAT16, FLOAT	ND
p	Optional attribute	Norm order. The value range is [0, +∞], and the default value is 2.0.	FLOAT	-

Constraint Description

• The data types of x1, x2, and y must be the same.
• The dimensions of x1 and x2 range from 2 to 8, and the lengths of the last dimensions must be the same.
• The dimensions of x1 and x2 except the last two dimensions must satisfy the broadcast relationship; the corresponding output dimensions are the broadcast shape.
• If the second-to-last dimension of x1 is P and the second-to-last dimension of x2 is R, the last two dimensions of the output are [P, R].
• p must be greater than or equal to 0. +∞ is supported, and common values are 0, 1.0, 2.0, and +∞.

Invocation Description

Invocation Method	Invocation Sample	Description
aclnn invocation	test_aclnn_cdist	Invokes the Cdist operator through the aclnnCdist API.