Marshall Lochbaum

You may also be interested in CBQN's current radix sort? For me it's coming up substantially faster than Julia, although surely this is somewhat architecture-dependent. Looking at `radix_sort` in sort.jl,...

CBQN doesn't have 64-bit ints: BQN specifies a single numeric type, which is double-precision floats for most implementations. So 32-bit ints and smaller types are an optimization. Like many things...

Maybe it would be better to do this filtering in `logical_merge`? That is, if one of the halves of an unsorted/not-unsorted pair has length less than the threshold, it should...

Thanks for answering the questions. They're just a sample of things I have to hop around files to understand; my point is that it would be much better to have...

Sorry, I'm frustrated as you can see. It is of course good work, which is why I'm taking the time to read it. The powersort integration and extension of branchless...

The script tests all combinations, so if `012 120 201` was the absolute best, it'd be in the table. I see that it's just a touch better than `000 111...

It occurred to me to try a narrower distribution, so I changed the 0, 4, 7 offsets to 1, 4, 6. It's a major improvement! Overall this distribution is much...

Outer offsets of 1, 3.5, 6 combined with inner offsets of 0, 3.5, 7 do better still, so the better thing to do, going by this model, is try to...

Random data I assume? I've been switching to [pisort](https://github.com/mlochbaum/SingeliSort/blob/master/src/merge.singeli#L134-L150) (simple merge, half of [piposort](https://github.com/scandum/piposort) more or less) at size 192 and below. Uses O(n) memory though.

There is a standard `expm1` function giving `exp` minus 1, so with that you could write `1-exp(-rate_of_life_formation)` as `-expm1(-rate_of_life_formation)` and avoid rounding to 0. In general, log odds-ratio representation gives...