Daniel Baker

That's a reasonable point. I think having separate `write` (for writing to a buffer) and `serialize` methods (for writing to a file) where the latter calls the former is probably...

I've updated this. `kh_read_##name` and `kh_write_##name` both work on `FILE *`s, while `kh_serialize_##name` and `kh_deserialize_##name` operate on paths.

It would be great if these calls could be inlined. I've been working with the [blaze](https://bitbucket.org/blaze-lib/blaze) linear algebra project to incorporate Sleef into its expression templates. It works currently, but...

I've used FFTW for C++ projects before, but was trying this to see if it was easier to distribute. Allocating enough memory isn't a problem now that I know, but...

Hi - I think I've patched it up now. Thanks - the error message was helpful in tracking down the problem. It should be fixed - I'll merge it soon,...

Hi again, It's been a long time, but I was finally able to track that down. That was a bug introduced when trying to add M1 support properly. I've patched...

Thanks for checking! It seems that for large sketches/small inputs, the default one-permutation setsketch approach is running into empty registers and something is going wrong when accounting for it. You...

Hi! That's right, the default addh uses Python's hash function which is seeded at each invocation. I should modify this to be consistent. In the short term, you can convert...

Yes, though that requires two passes through the data and with accompanying memory accesses. It's probably not that bad with compiler-optimized memcpy, though it would be nice to avoid if...

Thank you! Does FFHT assume that the memory has been aligned?