Research On HotSpot Profiling In Dynamic Reconfigurable Computing

As a new computing mode, Reconfigurable Computing is a trade off between performance and flexibility, and cover the gap between general CPU and ASIC. Reconfigurable Computing has not only high performance but also good flexibility. It is the third way between general computing and special computing.Based on the research of current achievements in Reconfigurable Computing, this thesis makes deep research on reconfigurable computing. The primary works of this thesis can be summarized as follows:(1) The examples of Reconfigurable Computing in academy and industry are carefully studied. Reconfigurable Computing's definition, conception, classification are also studied. A new design flow which aims at binary code is proposed. This design flow is better than the traditional in standardization, transparence, and indutrialization.(2) HotSpot profiling is studied. Trace selection algorithm evaluation is proposed, which is a quantitative method for the evaluation of Trace selection algorithm. Based on the analysis of NET algorithm, the improvement strategies are proposed.(3) CDFG contructing is studied. Analysis is made on IA-32codes, and the method for CDFG contructing is proposed. For the accumulation of CDFG contructing, an efficient chain table is presented.(4) Based on the proposed algorithm, a prototype system TBCC is proposed. Based on binary code, TBCC profiles HotSpot in the software, constructs CDFG of the HotSpot, and gets ready for dynamic reconfigurable computing. SPEC Int 2000 is used to test the HotSpot profiling algorithm. Performance analysis proves that the improved NET algorithm is better than previous NET algorithm. TBCC has the fast execution performance in CDFG contructing.