数据处理方法及装置、处理模组、终端设备及存储介质
By converting the convolution kernel into a second convolution kernel that is more suitable for the processing unit, and converting the data into suitable second data, the problem of wasted convolution computing resources is solved, and the data processing efficiency and the intelligence of terminal devices are improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOPHGO TECH LTD
- Filing Date
- 2022-11-14
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, convolutional computation can only utilize 3/32 or 3/64 of the processor's computing power, resulting in wasted processor resources, long data processing latency, poor intelligence of terminal devices, and poor user experience.
By determining whether the number of processing units is greater than the length of the first dimension of the convolution kernel, the first convolution kernel is converted into the second convolution kernel using the target transformation relationship, and the data to be convolved is converted into the second data. Convolution operations are performed using the processing units, making full use of the parallel processing mechanism of the processing units.
It improves data processing efficiency, makes full use of the computing power of the processing unit, reduces resource waste, and increases the speed of data processing and the intelligence of terminal devices.
Smart Images
Figure CN115906939B_ABST