Maximizing SIMD resource utilization in GPGPUs with SIMD lane permutation
Maximizing SIMD resource utilization in GPGPUs with SIMD lane permutation
Minsoo Rhu,M. Erez
2013 · DOI: 10.1145/2485922.2485953
International Symposium on Computer Architecture · 53 Citations
TLDR
SIMD lane permutation (SLP) is proposed as an optimization to expand the applicability of compaction in such cases of lane alignment, improving SIMD utilization up to 34% accompanied by a maximum 25% performance boost.