Single-instruction multi-thread mode oriented method for DMA transmission in GPDSP

Abstract

The invention discloses a single-instruction multi-thread mode oriented method for DMA (Director Memory Access) transmission in a GPDSP (General Purpose Digital Signal Processor). The method comprises: transferring data of SIMT (Single Instruction Multiple Thread) programs irregularly stored in an out-of-core storage space to a VM (Vector Memory) of a core by configuring a DMA transmission transaction once; and after the transfer, regularly storing the data in the VM, and enabling vector calculation parts to concurrently access to the data. The method is simple in principle, convenient to operate and configurable in data transmission amount, can efficiently supply data to the SIMT programs in a background operation manner, and not only better supports the execution of the SIMT programs but also greatly improves the operational performance of the GPDSP.

Description

technical field [0001] The present invention mainly relates to the field of direct storage access components (DirectorMemoryAccess, DMA) of a general purpose digital signal processor (GeneralPurposeDigitalSignalProcessor, GPDSP), in particular to a kind of DMA transmission method facing single instruction multiple thread (SingleInstructionMultipleThreads, SIMT) mode data requirements, using The data of the SIMT program irregularly stored in the external storage space is neatly moved to the internal vector storage unit (VectorMemory, VM) through the DMA transfer transaction, and finally it is convenient for each vector unit to execute in parallel. Background technique [0002] Digital Signal Processor (Digital Signal Processor, DSP), as a typical embedded microprocessor, is widely used in embedded systems. It has the characteristics of powerful data processing capability, good programmability, flexible use and low power consumption. It has brought great opportunities to the d...

AI Technical Summary

Problems solved by technology

[0004] General scientific computing requires high-performance DSP, but traditional DSP has the following disadvantages when used in scientific computing: 1) The bit width is small, which makes the calculation accuracy and addressing space insufficient

However, the current vector memory access operation only provides access to vector data with specific changing rules such as continuous address or equal address stride, which cannot meet the needs of multi-threaded parallel execution.

Images