High-concurrency sequence alignment calculation acceleration method based on CPU + GPU isomerism

Abstract

The invention discloses a high-concurrency sequence alignment calculation acceleration method based on CPU + GPU isomerism. The method comprises the following steps: reconstructing BWA-MEM algorithm codes; performing task concurrent processing on the CPU: completing division of a sequence set, and forming a plurality of concurrent tasks for the first time; running the BWA-MEM algorithm after code reconstruction, and completing concurrent processing of data on the GPU; and task concurrent processing on the GPU: for seed sets and chains generated in the sequence data comparison process, dividing the seed sets with the same or adjacent length, position and quantity into the same data block and chain, and performing the same processing, thereby completing the division of the seed sets and the chains, and forming a plurality of concurrent tasks for the second time. According to the method, the characteristics of the BWA-MEM algorithm and the characteristics of GPU acceleration equipment are closely combined by designing a task parallel and data parallel mode, the strong concurrent operation capability of the GPU is fully utilized, excellent performance is provided for a sequence alignment algorithm, and the efficiency of high-concurrent processing is higher.

Description

technical field [0001] The invention relates to the field of biological sequence comparison, in particular to a CPU+GPU heterogeneous high-concurrency sequence comparison calculation acceleration method. Background technique [0002] Biological sequence alignment is the application of the classic text alignment problem in the computer field to the biological field. With the development of emerging new molecular biology techniques, subsequent molecular biology research such as gene variation, RNA expression, protein and gene interaction requires researchers to use high-throughput methods to explain. This poses new challenges for high performance computing. The challenge in the era of high-throughput sequencing is no longer data generation, but data storage, processing and analysis. The development of third-generation sequencing technology will further accelerate the sequencing speed and generate longer sequencing fragments, which puts forward higher requirements for the dev...

AI Technical Summary

Problems solved by technology

[0006] The above technical solutions are to optimize or improve a specific algorithm (such as SW algorithm, FM-index algorithm). These technical solutions focus on the algorithm itself and use the characteristics of GPU to solve the problem of data parallelism. The improvement of the specific algorithm used to achieve the purpose of increasing the speed of biological sequence comparison is that its disadvantage is that when performing biological sequence comparison, it is necessary to find a relatively accurate matching sub-segment as a seed. For different reads, it contains The length, position, and number of SMEMs are very different; when using the task division method of processing one reads per thread on the GPU platform, this will cause serious desynchronization between different threads, including reads with shorter SMEMs It is necessary to wait for the reads containing longer SMEMs to be found, and for the reads containing shorter SMEMs, the number of SMEMs it contains is more, so the reads containing longer SMEMs need to wait for the reads containing shorter SMEMs in turn. This situation of waiting for each other leads to extremely low utilization of computing resources, and the acceleration effect is very limited

The above technical solutions focus on the problem of data parallelism (further increasing the ability of data parallelism), and cannot overcome the above shortcomings

[0007] Similarly, due to the seeding method of BWA-MEM that is highly related to the input data, the number and length of seeds contained in each reads will be very different, which will affect the process of generating chains based on seeds in the next module, resulting in the generation of The different threads of the chain module have poor synchronization, which further limits the acceleration effect

Since the operating characteristics of the algorithm are highly correlated with the input data, these methods cannot be suitable for input data with different characteristics at the same time

The requirement of the GPU platform for the consistency of instructions in the same warp severely limits the acceleration effect of the BWA-MEM algorithm.

Therefore, the original BWA-MEM algorithm also cannot solve the above problems

Images