Research And Implementation Of Hardware Prefetching Technology Based On BOOM

The CPU needs to continuously read and write the data from the cache in the modern computer architecture.However,with the continuous and rapid development of modern processor design technology and the improvement of manufacturing level of integrated circuit,the difference between the speed of read-write access to the cache and the calculation speed of the processor becomes more and more obvious.This restricts the development of the performance of the processor.The hierarchical memory system solved the problem.but it also brought a new problem.when the processor can`t tag the address in the cache,the request needs to arrive at the lower storage structure for the data.In this process,there is a long clock delay,it affects the improvement of processor performance.The prefetching technology enables the processor to get the relevant data before it generates the cache miss.In this way,the prefetching technology can achieve the purposes of hiding the time delay of the memory access and improving the performance of processors.The BOOM is a typical super scalar processor that is based on the RISCV.There is no design of related prefetch units in the BOOM processor,so there is much room for improvement of performance.This paper is based that BOOM as an experimental platform.It describes the design features of the related cell modules in the BOOM,such as L1 Dcache ? TLB and the LSU.In this paper,we design two methods in light with its character in design to prefetch data for L1 Dcache,which are sequential hardware prefetch and the prefetch based on virtual address and the stride reference.In addition,performance counters are added to the hardware design of the BOOM and the relevant allocation address is added to the compiler in the software for monitoring the performance parameters of the processor.Finally,we use the SPEC CPU 2006 to test the performance of BOOM processor based on the Xilinx zc706.By comparing the data of FPGA simulation and design principles,we analyze and summarize the advantages and disadvantages of the different methods.The experimental results show that the different methods has respectively realized the increased performance of the BOOM by 3% and 5%,they achieved the original purpose of the design.