Table 2 Summary of CUDAlign versions
From: DNA sequences alignment in multi-GPUs: acceleration and energy payoff
Version | Major contributions | Ref. |
|---|---|---|
1.0 | Uses SW with affine gap (Gotoh) to compare long sequences on GPUs. It ouputs the optimal score and the end coordinates of the optimal alignment. | [50] |
2.0 | Includes an adapted version of Myers-Miller to retrieve the optimal alignment in linear space. | [51] |
2.1 | Block pruning optimization. | [29] |
3.0 | Multi-GPU version for SW phase 1 (distributes the matrix, overlaps computations with communications). | [49] |
4.0 | Multi-GPU version for SW phases 1 and 2, including Incremental Speculative Traceback (IST) to accelerate the multi-GPU retrieval of the optimal alignment. | [30] |
MASA | Multi-platform architecture for sequence aligner, enabling versions to run on (1) multicore CPU using OpenMP, (2) multicore CPU using OmpsSs, (3) manycore GPU using CUDA, and (4) Xeon Phi using OpenMP. | [52] |