Daniel Lemire

Lift and/or document alignment requirements

1

Schemes like simdfastpfor have implicit alignment requirements. These alignment restrictions are neither necessary nor useful, and they are not documented. They have been lifted in the follow-up research library [SIMDCompressionAndIntersection](https://github.com/lemire/SIMDCompressionAndIntersection).

We have a vectorized encoder: https://github.com/lemire/streamvbyte

The current GeneralHash implementation supports 19-bit hash values, but nothing higher. It should be easy to support 32-bit or higher hash values. http://www.wolframalpha.com/input/?i=factor+x^0+x^1+x^2+x^5+x^6+x^7+x^9+x^11+x^14+x^19+x^20+x^21+x^22+x^24+x^27+x^30+x^32+mod+2 c.c. @vcunat

This is clever code that could be put to good use: char * s =... // null terminated string uint64_t c = *( uint64_t *) (s); uint64_t zero_mask = (c...

https://github.com/google/farmhash

https://github.com/google/highwayhash/ https://news.ycombinator.com/item?id=11212851 c.c. @jbapple

This library was used to develop and test Bloofi : https://github.com/lemire/bloofi (I am raising this issue merely to thank you for making the code available in the first place.)

http://xorshift.di.unimi.it/xoroshiro128plus.c

Optimize to_bitmask() for ARM

2

The current to_bitmask() implementation in ARM is expensive. In simdutf, we reengineered the code to use a faster function... https://github.com/simdutf/simdutf/blob/c0cb6dddb12256a261bb249255660b79c1fe67a2/src/simdutf/arm64/simd.h#L210-L212 Reference: https://community.arm.com/arm-community-blogs/b/infrastructure-solutions-blog/posts/porting-x86-vector-bitmask-optimizations-to-arm-neon We should investigate this optimization in simdjson. Credit...