Robert Burke

Results 32 comments of Robert Burke

@alreadydone thanks! I asked in large part because my profiler told me I was spending a ton of time regenerating Dirichlet noise for each search after copying from someone who...

sum can overflow on legal inputs. also, looks like this is ~100% faster than that other submission.

Yes, you can assume that byte-wise different strings are different. Because normalization is about equivalence of sequences of code-points and because the reference solution does not contain any code to...

I think "vxsort and djbsort are faster for large inputs" is not a claim that accompanies the benchmark. I think a claim that accompanies the benchmark is that the small...

If some techniques are slow for inputs in a some ranges and fast for inputs in different ranges, maybe they can be applied only to ranges for which they are...

![annotated graph dividing vqsort vs competitors cycles-per-byte into regions based on input size and performance](https://cdn.discordapp.com/attachments/606246764651085829/988375089936490526/FUr3FGGUYAAjOua.png) If you can forgive the mess I've made here, I think the situation is like...

Super cool! It sounds plausible to me that the extra branches turn out to be too costly and that the sadness in region C was actually due in large part...

I tried to reproduce the benchmark but I do not have any Skylake machines and results on Haswell seem less weird than the results djb posted.