Figure 4

Organization of threads in a block of the modified SW alignment function. The modified SW algorithm is performed for a group of 32 query sequences in the x direction in parallel. As the query length is limited up to 32 nucleotides, 32 independent cells can be calculated at a time using 8-thread parallel update of groups of 4 nucleotides in the y direction. The above thread organization scheme makes sure that all 32 threads with in the same warp can operate on the same nucleotide location, which is able to minimize the intra-warp branch and bank conflict in shared memory.