Spence

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 Spence function (dilogarithm) value of each element in the input tensor and outputs the result.

• Computation formula:

$$\text{Spence}(x)={\int }_1^x\frac{\ln t}{1-t}dt={\text{Li}}_2(1-x)$$

Special values:

• Outputs NaN when $x<0$
• Outputs ${\pi }^2/6\approx 1.6449$ when $x=0$
• Outputs $0$ when $x=1$
• Outputs NaN when $x=+\infty$

Parameter Description

Parameter Name	Input/Output/Attribute	Description	Data Type	Data Format
x	Input	Independent variable of the Spence function, corresponding to x in the formula. Elements with x < 0 output NaN.	FLOAT16, FLOAT, BFLOAT16	ND
y	Output	Computation result of the Spence function, corresponding to Spence(x) in the formula.	FLOAT16, FLOAT, BFLOAT16	ND

Constraint Description

• FLOAT16 and BFLOAT16 inputs are promoted to FLOAT32 for computation inside the kernel, and the results are converted back to the original data type.
• The number of input dimensions must not be greater than 8; otherwise, the operator reports an error. Scalar inputs (processed as a Tensor with one element), empty tensors, dynamic shapes, and dynamic ranks are supported.
• Non-contiguous inputs are automatically made contiguous by the framework before computation.

Invocation Description

Invocation Method	Sample Code	Description
GE graph mode	test_geir_spence	Invokes the Spence operator in GE graph mode. The prototype file that defines the operator IR is spence_proto.h.