Math API Sample Introduction
项目地址: https://gitcode.com/cann/asc-devkit Overview
This sample collection demonstrates typical usage of various Math API features and provides corresponding end-to-end implementations.
Sample List
| Directory Name | Description | Supported Products |
|---|---|---|
| acosh | This sample demonstrates an operator implementation based on the Acosh high-level API. The sample computes the inverse hyperbolic cosine function element-wise | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| axpy_half_float | This sample implements the functionality of multiplying each element in the source operand src by a scalar and adding it to the corresponding element in the destination operand dst using Axpy. The data type combinations for the source and destination operands of the Axpy interface can only be: (half, half), (float, float), or (half, float). In this sample, the input tensor and scalar data types are half, and the output tensor data type is float. This sample implements the Axpy operator using Ascend C and uses the <<<>>> kernel call operator to complete the basic flow of running and verifying the operator kernel function on the NPU side, providing the corresponding end-to-end implementation | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| bitwiseand | This sample demonstrates an operator implementation based on the BitwiseAnd high-level API. The sample performs a bitwise AND operation on two inputs | Ascend 950PR/Ascend 950DT |
| ceil | This sample demonstrates an operator implementation based on the Ceil high-level API. The sample obtains the smallest integer greater than or equal to x, that is, rounding toward positive infinity | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| clamp | This sample demonstrates an operator implementation based on the Clamp high-level API. Values in the input that are not nan and fall outside the [min, max] range are clipped to max or min. If min is greater than max, all non-nan values are set to max | Ascend 950PR/Ascend 950DT |
| clampmin | This sample demonstrates an operator implementation based on the ClampMin high-level API. The sample replaces numbers in srcTensor that are less than scalar with scalar, while numbers greater than or equal to scalar remain unchanged, and outputs as dstTensor | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| cumsum | This sample introduces how to call the CumSum high-level API to implement the cumsum operator, which computes cumulative sums along rows or columns of an input tensor | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| erf | This sample demonstrates an operator implementation based on the Erf high-level API. The sample computes the error function element-wise | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| exp | This sample demonstrates an operator implementation based on the Exp high-level API. The sample computes the natural exponential element-wise. Users can choose whether to use the Taylor series expansion formula for computation | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| fma | This sample demonstrates an operator implementation based on the Fma high-level API. The sample computes the result of multiplying two inputs element-wise and then adding a third input | Ascend 950PR/Ascend 950DT |
| fmod | This sample demonstrates an operator implementation based on the Fmod high-level API. The sample computes the remainder of dividing two floating-point numbers a and b element-wise | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| frac | This sample demonstrates an operator implementation based on the Frac high-level API. The sample computes the fractional part element-wise | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| isfinite | This sample demonstrates an operator implementation based on the IsFinite high-level API. The sample determines whether input floating-point numbers are neither NAN nor INF element-wise. The output result is either a floating-point number or a boolean value | Ascend 950PR/Ascend 950DT |
| log | This sample demonstrates an operator implementation based on the Log high-level API. The sample computes logarithms with base e, 2, or 10 element-wise | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| logicaland | This sample demonstrates an operator implementation based on the LogicalAnd high-level API. The sample performs a logical AND operation element-wise. When the input data type is not bool, zero is treated as False and non-zero data is treated as True | Ascend 950PR/Ascend 950DT |
| power | This sample demonstrates an operator implementation based on the Power high-level API. The sample implements element-wise power operations and supports three modes: tensor-to-tensor, tensor-to-scalar, and scalar-to-tensor power operations | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
| rint | This sample demonstrates an operator implementation based on the Rint high-level API. The sample obtains the integer closest to the input data. If two integers are equally close, the even number is selected | Ascend 950PR/Ascend 950DT |
| where | This sample demonstrates an operator implementation based on the Where high-level API. The sample selects elements from two source operands based on specified conditions to generate a destination operand. Both source operands can be either LocalTensor or scalar | Ascend 950PR/Ascend 950DT |
| xor | This sample demonstrates an operator implementation based on the Xor high-level API. The sample performs XOR operations element-wise | Ascend 950PR/Ascend 950DT Atlas A3 Training Series Products/Atlas A3 Inference Series Products Atlas A2 Training Series Products/Atlas A2 Inference Series Products |
创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
