It seems like the project has been abandoned? Last commit a year ago.

Check out CozoDB, the embedded datalog-queried hybrid relational+vector+graph database written in Rust: <a href="https:&#x2F;&#x2F;www.cozodb.org&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.cozodb.org&#x2F;</a>I used it in a toy application and it was awesome.This appears to be a dream database from the future.

I guess SQL is a requirement if they want to market their technology to normies.

I have some local-first&#x2F;client-side applications using datascript in ClojureScript. Used datahike (FOSS Datomic alternative) some times on the backend too, but mostly tend to use XTDB nowadays, which used to have a Datalog API but I think they removed in favor of SQL-like way instead, which was kind of a shame.

Would like to hear about the ergonomic problems you have with souffle. We integrate it into our rust tools quite well, and generate typesafe rust bindings to our souffle programs, allowing us to insert facts and iterate over outputs.

Just gave a talk about this: <a href="https:&#x2F;&#x2F;blog.waleedkhan.name&#x2F;what-if-sql-were-good&#x2F;" rel="nofollow">https:&#x2F;&#x2F;blog.waleedkhan.name&#x2F;what-if-sql-were-good&#x2F;</a>- Recommend Ascent (Rust only, but supports targeting WASM)- Soufflé: good, but too hard to integrate into existing systems; lots of ergonomic problems in comparison to Ascent (can elaborate)- CozoDB: really cool but seems to be abandoned- Logica: have not tried it yet

For a while the Rust compiler&#x27;s borrow checker &quot;Polonius&quot; was implemented with datalog using the `datafrog` engine. However, it appears to me that the in-tree version of polonius is moving away from datafrog (not enough of a rustc expert to say for sure which version of the borrow checker engine is in use)

CodeQL compiles to the Souffle datalog engine and I use it for static analysis. I&#x27;ve also used ascent for a few random side projects in Rust which is very convenient.

On a side note, what tools that leverage Datalog are in use by the HN crowd?I know that Datomic[0] is very popular. I&#x27;ve also been playing with Clingo[1] lately.[0] <a href="https:&#x2F;&#x2F;www.datomic.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.datomic.com&#x2F;</a>[1] <a href="https:&#x2F;&#x2F;potassco.org&#x2F;clingo&#x2F;" rel="nofollow">https:&#x2F;&#x2F;potassco.org&#x2F;clingo&#x2F;</a>

The work done&#x2F;supervised by Kristopher Micinski on using HPC hardware (not only GPUs but clusters) for formal methods is really encouraging. I hope we reach a breakthrough of affinity between COTS compute hardware and all kinds of formal methods, as GPUs found theirs with deep learning and subsequent large models.One possible answer to &#x27;what do we do with all the P100s, V100s, A100s when they&#x27;re decomissionned from their AI heyday (apart from &#x27;small(er) models&#x27;.

(have been a big fan of this work for years now)From the nearby perspective of building GFQL, an embeddable oss GPU graph dataframe query language somewhere between cypher and duckdb&#x2F;pandas&#x2F;spark, at an even higher-level on top of pandas, cudf, etc:It&#x27;s nice using higher-level languages with rich libraries underneath so we can focus on the foundational algorithm &amp; data ecosystem problems while still achieving crazy numberscudf gives us optimized GPU joins, so jumping from cheap personal CPU or GPU boxes to 80GB server GPUs and deep 2B edge whole-graph queries running in a second without work has been nice :) we want our focus on getting regular graph operations fully data parallel in the way we want while being easy for users, figuring out areas like bigger-than-memory and data lakes, etc, so we want to defer lower-level efforts to when the rust etc rewrite is more merited. I do see value in starting low when the target value and workload is obvious for building our (eg, vector indexes &#x2F; DBs), but when breaking new ground at every point, value to going where you can roll &amp; extend faster.

From their publication history, they want to use all HPC niceties, to use most&#x2F;any available HPC installations.Nowadays that means mostly CUDA on NVIDIA and HIP on AMD on the device side. Curious how the spirv support is on NVIDIA GPUs, including nsight tooling and the maturity&#x2F;performance of libraries available (if only the cub-stuff for collective operations).

Just depends on how the manufacturer of the GPU handles code written in different languages. For example, what level of API access, what level of abstraction, and how is the source compiled i.e. how optimized is it. For example, on an apple GPU, you’ll see benchmarks that openCL and metal can vary depending on the tasks.

Why is cuda sub-optimal compared to SPIR-V? I don&#x27;t think I know the internals enough to understand if it&#x27;s supposed to be obvious why one is better than the other.I&#x27;m currently sitting and learning cuda for ML purposes, so happy to get more educated :)

What kind of SPIR-V? The SPIR-V used for compute shaders (Vulkan Compute) is totally different to the one for compute kernels (OpenCL and SYCL)...

Curious, why use cuda and hip? These frameworks are rather opinionated about kernel design, they seem suboptimal for implementing a language runtime when SPIR-V is right there, particularly in the case of datalog.

Optimizing Datalog for the GPU