Expint
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 exponential integral Ei(x) = PV ∫_{-∞}^{x} (eᵗ/t) dt (Cauchy principal value integral). The computation formula is:
Ei(x)=PV∫−∞xettdtEi(x) = PV \int_{-\infty}^{x} \frac{e^{t}}{t} \mathrm{d}t
Ei(x)=PV∫−∞xtetdt
Parameter Description
| Parameter Name |
Input/Output/Attribute |
Description |
Data Type |
Data Format |
| x |
Input |
Input parameter for the exponential integral computation, which is x in the formula. |
BFLOAT16, FLOAT16, FLOAT |
ND |
| y |
Output |
Output parameter of the exponential integral computation, which is Ei(x) in the formula. |
BFLOAT16, FLOAT16, FLOAT |
ND |
Constraint Description
- The DOUBLE (FP64) data type is not supported.
- When the input x < 0, the output is NaN; when x = 0, the output is -inf; and when x = +inf, the output is NaN.
- For x > 0, the output is the finite true value of Ei(x); for x > ~88.72, the output saturates to +inf due to fp32 exp(x) intermediate overflow (consistent with the third-party reference).
- The shape and data type of the output are inferred by the framework based on the input (the computation is element-wise, and the output is the same as the input x), and the data format is ND.
- 0- to 8-dimensional ND tensors are supported, including scalars, empty tensors, dynamic shapes, and dynamic ranks. Tensors with 9 or more dimensions are not supported.
- A non-contiguous tensor input is automatically converted to a contiguous tensor by the framework before execution.
Invocation Description
| Invocation Method |
Invocation Sample |
Description |
| Graph-mode invocation (static shape) |
test_geir_expint |
Invokes the Expint operator through the graph mode. |
| Graph-mode invocation (dynamic shape) |
test_geir_expint_dynamic |
Invokes the Expint operator through the graph mode, supporting scenarios with unknown dimensions (-1) and unknown ranks (-2). |