基于程序执行轨迹的SoC硬件加速模块提取方法

程序执行轨迹（Program executions trace，以下简称trace）是程序执行过程的指令流信息的记录，trace完整地记录了程序执行过程中所执行指令的内容和顺序。对于大多数程序，少数几个较短的热trace决定了系统的总体性能。本文提出了基于程序执行轨迹提取加速模块的软硬件划分方法。利用热trace提取算法划分系统中关键的trace到硬件，使用分支断言构造原子执行单位，以较小的硬件代价获得较高的加速比。在本文实验中，与采用模拟退火算法的指令级细粒度划分相比，获得的性能平均高9．6％，最终结果硬件面积小29％。

A Method to Construct SoC Hardware Acceleration Module with Trace

Traces are the records of the program executions, and for most programs a few of short hot traces affect the overall performance. In this paper, we propose a hardware/software partitioning method to extract the hardware acceleration modules based on traces. With the algorithm presented, hot traces are identified and transformed to atomic execution units using branch assertions to get high speedup at low cost. Experiment result shows that compared with annealing partitioning in granularity of instruction, the trace based method produces higher performance by average 9.6% and the hardware area is reduced by 29%.