Figure 3.

Adaptive tile-based strategy. The optimal tile size T can be calculated according to the hardware configuration of GPU. After calculated, the T subject sequences together are transferred to GPU global memory. Permutations and alignments are done in parallel in GPU. Then T × 1000 alignment scores are moved back to CPU for T fittings.

Zhang et al. BMC Bioinformatics 2012 13(Suppl 5):S3   doi:10.1186/1471-2105-13-S5-S3