Accelerator execution method and electronic equipment

Abstract

A method performed by an accelerator and an electronic device are described herein. The method comprises the following steps: receiving a first tensor multiplication instruction for a first thread of an accelerator; the first thread set broadcasts a second set of factors in the second tensor to the second thread set based on a memory logical address for the second tensor; a first thread in a second set of threads dot-product operates a first set of factors and the second set of factors based on the first factor register representation to generate a first dot-product set in a first row of a third tensor. According to the method, the matrix is decomposed, and the threads are distributed according to the rows, so that the multiple threads can process multiple rows of the matrix tensor in parallel, and the processing efficiency of matrix multiplication is improved. Besides, programming personnel know the row and column structure of the matrix tensor and the thread condition in the accelerator during programming, so that the threads can be flexibly used for processing matrix multiplication in parallel, and the programming flexibility is improved.

Description

technical field [0001] Embodiments of the present disclosure relate generally to the field of electronics, and more particularly to a method performed by an accelerator and an accelerator. Background technique [0002] Parallel high-performance multi-threaded multi-core processing systems such as graphics processing units (GPUs) can process data much faster than in the past. These processing systems can break down complex computations into smaller tasks and are processed in parallel by multiple cores to increase processing efficiency and reduce processing time. [0003] In some cases, multi-core processors such as GPUs are particularly advantageous for processing tensors with large amounts of data in the same or similar form. In the computer field, tensor data usually represents data of one-dimensional or multi-dimensional arrays. For example, image data is a conventional two-dimensional tensor data, which can be represented by two-dimensional arrays. For another example, ...

AI Technical Summary

Problems solved by technology

Conventional matrix multiplication based on internal accelerators such as GPUs is usually not known to program programmers, so programmers usually do not understand the process of hardware performing matrix multiplication, and thus cannot optimize the calculation of matrix multiplication for hardware, which leads to program execution Efficiency and generally less efficient tensor processing

Images