Research On BWA-MEM Core Algorithm Based On GPU Acceleration

Advances in genetic technology have a profound impact on many fields such as pharma-ceutical research and development,personal health,and so on,and therefore have spawned a large number of important related technologies.Gene comparison is one of them.As a com-bined field of genetic engineering technology and computer technology,gene comparison technology mainly uses computing equipment to match a large number of disordered gene sequences to the standard genome,providing key data for analyzing biological characteris-tics.This is of great significance for studying the genetic characteristics of biological indi-viduals,races,and communities,and has a profound impact on the exploration of unknown biological information.This article takes Illumina's gene sequencing solution as the basic standard,takes base sequences of 100bp-200bp in length as the object,and conducts research based on BWA-MEM algorithm and GPU heterogeneous computing platform.With the improvement of detection technology,the scale of individual genetic data has in-creased,which makes the comparison algorithm have higher requirements for timeliness.Because traditional algorithms are difficult to meet the requirements of rapid personal de-tection,which restricts the further application of related technologies,this paper proposes a GPU-based parallel acceleration solution.The main contradiction that restricts the com-parison speed is in two aspects:the limitation of CPU computing power and the inherent redundancy of BWA-MEM algorithm.Focusing on these two issues,this paper proposes algorithm optimization logic and imple-mentation of parallel solutions by analyzing the hardware and software structures involved in GPU programming.Through CPU and GPU hybrid heterogeneous programming,the algo-rithm is accelerated in parallel between tasks and within tasks to improve efficiency.There are two main aspects of completing heterogeneous programming:First,realize inter-task parallelism in the inter task dimension.According to the principle of gene sequence consistency,by increasing the size of parallel processing tasks,reducing the number of memory accesses between the CPU and GPU,in order to ensure the merged memory access between tasks,by designing a stream processing scheme and a flattened data structure,different calculations are realized The combined access operation of resources to data.At the same time,intra-task parallelism is achieved in the intra task dimension.In-task par-allelism is mainly divided into two parts:exact match algorithm(exact match algorithm)and inexact match algorithm(inexact match algorithm).The core of the precise comparison algorithm is the SMEM algorithm.On the one hand,optimization reduces algorithm com-plexity through algorithm-level optimization,and on the other hand,improves efficiency through fine-grained parallel pairs.The core of the non-precision comparison algorithm is the S-W algorithm,which realizes the parallelization of the algorithm within the warp through asynchronous startup,and rationally uses high-speed memory access to improve the performance of the algorithm.In this paper,the parallel implementation model of BWA-MEM algorithm is designed by way of inter-task and intra-task.On the one hand,it studies heterogeneous programming on GPU devices,on the other hand,it completes the software design through the unified com-puting architecture(CUDA).Through the mutual consideration of hardware and software resources,the comparison performance of the algorithm is effectively improved.Finally,the experiment proves that the parallel acceleration effect is remarkable.