[ognarb]

> History/Motivations This project started as an exploration of using AI agents for software development. Based on experience tuning systems using Abseil's B+tree, I was curious if performance could be improved through SIMD instructions, a customized allocator, and tunable node sizes. Claude proved surprisingly adept at helping implement this quickly, and the resulting B+tree showed compelling performance improvements, so I'm making it available here.

It seems the code was written with AI, I hope the author knows what he is doing. Last time I tried to use AI to optimize CPU-heavy C++ code (StackBlur) with SIMD, this failed :/

[klaussilveira]

Both Codex/Claude Code are terrible with C++. Not sure why that is, but they just spit out nonsense that creates more work than it helps me.

Have you tried to do any OpenGL or Vulkan work with it? Very frustrating.

React and HTML, though, pretty awesome.

[inetknght]

On the other hand, I've been using Claude Code for the past several months at work in several C++ projects. It's been fine at understanding C++. It just generates a lot of boilerplate, doesn't follow DRY, and gets persnickety with tests.

I've started adding this to all of my new conversations and it seems to help:

    You are a principal software engineer. I report to you. Do not modify files. Do not write prose. Only provide observations and suggestions so that I can learn from you.

My question to the LLM then follows in the next paragraph. Foregoing most of the LLM's code-writing capabilities in favor of giving observations and ideas seems to be a much better choice for productivity. It can still lead me down rabbit holes or wrong directions, but at least I don't have to deal with 10 pages of prose in its output or 50 pages of ineffectual code.

[simonw]

Which models?

It's possible Opus 4.5 and GPT-5.2 are significantly less terrible with C++ than previous models. Those only came out within the past 2 months.

They also have significantly more recent knowledge cut-off dates.

[DrBazza]

In what scenarios are they terrible? I hope not every scenario. I've found Codex adequate for refactoring and unit tests. I've not used it in anger to write any significant new code.

I suppose part of the problem is that training a model on publicly available C++ isn't going to be great because syntactically broken code gets posted to the web all the time, along with suboptimal solutions. I recall a talk saying that functional languages are better for agents because the code published publicly is formally correct.

[FpUser]

I use ChatGPT with C++ but in very limited manner. So far it was overall win. I watch the code very closely of course and usually end up doing few iterations (mostly optimizing for speed, reliability, concurrency).

Also to generate boilerplate / repetitive.

Overall I consider it a win.

[nurettin]

I use Claude to generate C++ 23, it usually performs well. It takes a bit of nudging to avoid repeating itself, reusing existing functionality, not altering huge portions without running tests, etc. But generally it is helpful and knows what to do.

[seg_fault]

I had the same experience. C++ doesn't even compile or I have to tell it all the time "use C++23 features". I tried to learn OpenGL with it. This worked out a bit, since I had to spot the errors :D

[leopoldj]

I apologize if this is common knowledge. Modern C++ coding agents need to have a deep semantic understanding of the external libraries and header files. A simple RAG on the code base is not enough. For example, GitHub Copilot for VS Code and Visual Studio uses IDE language services like IntelliSense. To that extent, using a proper C++ IDE rather than a plain editor will improve the quality of suggested code. For example, if you're using VS Code, make sure the C/C++ Extension Pack is installed.

[LoganDark]

Oh hey, I wrote a Stackblur implementation in Rust. The trick I used is to SIMD across multiple rows/columns of the image rather than trying to SIMD the algorithm itself.

https://github.com/logandark/stackblur-iter

[shihab]

I'd love to see a breakdown of what exactly worked here, or better yet, PR to upstream Abseil that implements those ideas.

AI is always good at going from 0 to 80%, it's the last 20% it struggles with. It'd be interesting to see a claude-written code making its way to a well-established library.