highway
highway copied to clipboard
benchmark SIMD with highway
SIMD benchmark is hard to find if there is any, as each arch needs some hand crafting for the best result.
can we make a highway-based SIMD benchmark that runs on AVX|SVE|RVV to compare their performance apple to apple?
Hi @laoshaw , thanks for reaching out. This is a very interesting topic. We do have such a benchmark for vectorized quicksort, see hwy/contrib/sort/bench_*. We've tested this on x86, NEON and SVE, with interesting results (https://onlinelibrary.wiley.com/doi/abs/10.1002/spe.3142).
We'd welcome pull requests for benchmarks of other algorithms and I'm happy to help or advise on that if you would like to develop one.
Closing, feel free to reopen if you'd like to continue the discussion.