On-chip scratch-pad memory (SPM) management method facing multitasking embedded system

Abstract

The invention discloses an on-chip scratch-pad memory (SPM) management method facing a multitasking embedded system. The on-chip scratch-pad memory management method comprises the steps as follows: pre-analyzing a program code segment; tracking a program to obtain an internal storage instruction accessing sequence of the program, so as to acquire the accessing times of an internal storage code block and Cache miss information, and accounting and recording a space-time conflict set of the internal storage code block under access of the Cache; selecting an algorithm as required to obtain an optimized SPM distribution scheme; generating a code distribution dispersion loading file, newly mapping and distributing the program code segment, and newly compiling codes to obtain an optimal execution result. Via comprehensive consideration of accessing frequency, cache miss frequency, intertask and intra-task conflict, the required optimal distribution is obtained, the utilization rate of the SPM is maximized, and finally under the premise of ensuring the real-time performance of the program, an optimal execution time scheme or an optimal energy conservation scheme is obtained.

Description

Technical field: [0001] The invention belongs to the field of embedded real-time systems, in particular to an on-chip scratch pad memory management method oriented to multi-task embedded systems. Background technique: [0002] In the development process of embedded systems, since the development speed of the main memory has been much slower than that of the central processing unit, the low reading speed and high energy consumption of the main memory have caused it to become the bottleneck of the performance and energy consumption of many embedded systems. On-chip memory bridges this growing gap between main memory and CPU speed. [0003] In embedded systems, on-chip memory mainly includes two types of scratch pad memory (SPM, Scratch Pad Memory) and cache (Cache). The scratch pad memory SPM and the cache cache are essentially a kind of static random access memory (SRAM, Static Random Access Memory), and the access speed is very fast, close to the speed of the CPU. Cache is...

AI Technical Summary

Problems solved by technology

However, the current research is mainly focused on the system with only the scratch pad memory SPM, and the use of the scratch pad memory SPM+cache Cache (such as figure 1 ) There are relatively few studies on multitasking systems with storage systems

In the only optimization algorithm for the multi-task system of the scratch pad memory SPM+cache Cache storage system, the algorithm uses functions as the basic allocation unit, and only considers the energy consumption reduction obtained by putting a single function into the SPM. For multi-task systems Inter-task conflicts and conflicts between functions within tasks are not considered, and these conflicts between tasks and within tasks have a great impact on system performance and energy consumption

[0005] Through the tracking research on the execution process of multiple programs, it is found that in many programs, the functions with high access frequency or high cache Cache miss frequency are not necessarily the functions that cause the most inter-task and intra-task conflicts (such as figure 2 , among which the most visited frequency is A 0 and B 1 , the one with the highest miss frequency is A 0 , but the one that causes the most inter-task and intra-task conflicts is B 0 ), the existing technology often does not take into account access frequency, Cache miss frequency, and inter-task and intra-task conflicts, let alone make full use of SPM

[0006] Chinese patent (application number: 201310042340.6, patent name: fine-grained data allocation method for embedded on-chip heterogeneous memory), although this patent uses the linear programming method to solve the data allocation problem in the note-type memory SPM, but 1) it It is to discuss the problem of data allocation in the sticky note memory SPM, and the data allocation is very different from the code allocation, because the correlation between the codes in the multi-tasking system cannot be allocated in units of memory blocks, 2) it is Use the linear programming method to solve the SPM allocation problem, but due to the large number of variables and constraints, the calculation time complexity of the linear programming method is exponential and will consume a lot of time, which cannot meet the requirements of real-time systems at all.

Images