crepe VS highway

https://github.com/topics/datalog?l=rust ... Cozo, Crepe

Crepe: https://github.com/ekzhang/crepe :

> Crepe is a library that allows you to write declarative logic programs in Rust, with a Datalog-like syntax. It provides a procedural macro that generates efficient, safe code and interoperates seamlessly with Rust programs.

Looks like there's not yet a Python grammar for the treeedb tree-sitter: https://github.com/langston-barrett/treeedb :

> Generate Soufflé Datalog types, relations, and facts that represent ASTs from a variety of programming languages.

Looks like roxi supports n3, which adds `=>` "implies" to the Turtle lightweight RDF representation: https://github.com/pbonte/roxi

FWIW rdflib/owl-rl: https://owl-rl.readthedocs.io/en/latest/owlrl.html :

> simple forward chaining rules are used to extend (recursively) the incoming graph with all triples that the rule sets permit (ie, the “deductive closure” of the graph is computed).

ForwardChainingStore and BackwardChainingStore implementations w/ rdflib in Python: https://github.com/RDFLib/FuXi/issues/15

Fast CUDA hashmaps

Gdlog is built on CuCollections.

GPU HashMap libs to benchmark: Warpcore, CuCollections,

https://github.com/NVIDIA/cuCollections

https://github.com/NVIDIA/cccl

https://github.com/sleeepyjack/warpcore

/? Rocm HashMap

DeMoriarty/DOKsparse:

Inspired by past work: LINQ, inline-python, crepe, DataScript, Riffle.

I quite like highway.

As mentioned, last time I tried vqsort for RVV it was surprisingly slow.

I tried to replicate it yesterday, but noticed that vqsort is now disabled for RVV: https://github.com/google/highway/blob/400fbf20f2e40b984be12...

Does highway support sorting networks for non-128-bit vector registers?

When I tried to compile it for AVX512, the BaseCase seems to only use xmm registers: https://godbolt.org/z/qr9xoTGKn

I'm not proficient in Rust, but API wise I'd conceptually define types like f32xn or f32s, which have the number of elements that fit into a vector register for your target architecture, so 4 for NEON/SSE, 8 for AVX and 16 for AVX512.

I can recommens lookig at the highway library: https://github.com/google/highway

What about Highway? https://github.com/google/highway I suppose that's C++ not C though.

google has a mature C++ library for portable SIMD. The original article seems to be a translation of the excellent algorithmica site which had it in C++.

https://github.com/google/highway

> Or do they somehow adapt to the operations supported by the user's CPU?

This is called runtime dispatch. You can do it manually or use a library, like Google Highway. GCC supports multiversioning where you write separate versions of a function and the right one is selected at runtime.

https://github.com/google/highway

https://gcc.gnu.org/onlinedocs/gcc-9.1.0/gcc/Function-Multiv...

GitHub

It's great you bring up cmp, helps to understand why 4x128 is not necessarily as good as 1x512. Quicksort, hardly a 'weird kernel', does comparisons followed by compaction. Because comparisons return a predicate, which have only a single write port, we can only do 128 bits of comparisons per cycle. Ouch.

However, masking can still help our VQSort [1], for example when writing the rightmost partition right to left without stomping on subsequent elements, or in a sorting network, only updating every second element.

[1] https://github.com/google/highway/tree/master/hwy/contrib/so...