Threadwell

Recent Notes

NOTimothyLottes
Aug 25, 2026
cmuratori
Aug 25, 2026
Threads
Aug 25, 2026
- archive

nicbarkeragain

72 items with this tag.

Aug 25, 2026
nicbarkeragain
Aug 05, 2026
An example of this - a classic way to construct a tree in code is with push / pop style calls.
Aug 03, 2026
IMO basically every "object creation API" you have should take a string label that is for internal use only.
Jul 15, 2025
I can't help but have this feeling that we're missing something in programming language design that will make SIMD much easier for the average programmer to use.
Aug 08, 2026
@nicbarkeragain This just saved me from hours of debugging.
Aug 06, 2026
Rust's ownership + RAII makes such patterns really easy and nice to encode statically, no debug strings needed if the compiler guarantees you can't open another object at the same level if the previous one wasn't closed.
Aug 05, 2026
@nicbarkeragain Thanks a lot for the example
Aug 05, 2026
@nicbarkeragain 1.
Aug 05, 2026
@nicbarkeragain Ah, matching strings, just like xml 🙃
Aug 05, 2026
@nicbarkeragain DeferLoop is super useful for this
Aug 05, 2026
@nicbarkeragain this particular case though is easily solved with `defer` or even `closures`
Aug 05, 2026
I love everything about this.
Aug 05, 2026
@nicbarkeragain That's a great tip.
Aug 05, 2026
@nicbarkeragain Or you could use C++ with RAII and not have to tie yourself in knots
Aug 05, 2026
@nicbarkeragain tangently, this example illustates pretty good why I like syntax with `end` keyword and optional name of closee better than just `{ ...
Aug 04, 2026
@nicbarkeragain I accidentally discovered this myself.
Aug 04, 2026
@nicbarkeragain function "draw string, for debug, in my game's 3d graphical space"
Aug 04, 2026
@nicbarkeragain WebGPU has this and it’s a life saver for debugging
Aug 03, 2026
@nicbarkeragain I do and access it via 9p.
Aug 03, 2026
@nicbarkeragain brb adding this to my codebase
Aug 03, 2026
@nicbarkeragain You can also implicitly pass the source code location from creation so there's no need for explicit labeling a lot of the time.
Aug 03, 2026
@nicbarkeragain This is the way!
Aug 03, 2026
@nicbarkeragain I don't suppose you could demonstrate this with a small code snippet/example that shows HOW one can label an allocation?
Aug 03, 2026
@nicbarkeragain Isn't this just a Python-style docstring with extra steps?
Aug 03, 2026
@nicbarkeragain https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/wdm/nf-wdm-exallocatepool2 🤘
Aug 03, 2026
@nicbarkeragain Can you post an example of how you would implement it?
Aug 03, 2026
@nicbarkeragain IIRC sokol_gfx does this.
Aug 03, 2026
@nicbarkeragain When I've seen this they usually require unique names and then you end up recapitulating the entire creation path in the string to make it unique but still readable, which I kinda hate.
Aug 03, 2026
@nicbarkeragain That's actually a smart idea.
Aug 03, 2026
@nicbarkeragain ya tagging allocations, and frames to actual assets in-game is absolutely necessary for development once your game stops being a small thing
Jul 16, 2025
@nicbarkeragain What popped into my head is languages like Uiua
Jul 16, 2025
@nicbarkeragain But is it really that hard to use though? Or what scenario do you mean? If we are to make use of SIMD we will be doing operations on collections of numbers.
Jul 16, 2025
@nicbarkeragain Hmmn.
Jul 16, 2025
@nicbarkeragain I’ve been wondering about this for over twenty years, but never found anything I felt was truly accessible.
Jul 16, 2025
@nicbarkeragain Honestly it will definitely be a random idea someone has at 3 am and everyone will just go "how did we not think about that earlier?" and then a revolution begins
Jul 15, 2025
@nicbarkeragain The answer exists.
Jul 15, 2025
@nicbarkeragain Reduced instruction sets.
Jul 15, 2025
@nicbarkeragain ISPC
Jul 15, 2025
@nicbarkeragain Mojo lang has SIMD aware data types and makes it a first class citizen.
Jul 15, 2025
My gamedev head just wants a sort of game mechanic lego for conveying a simd operation frame.
Jul 15, 2025
@nicbarkeragain I have hope that this is true.
Jul 15, 2025
@nicbarkeragain FFmpeg thinks otherwise.
Jul 15, 2025
@nicbarkeragain Have you ever studied APL?
Jul 15, 2025
@nicbarkeragain 128-bit seems to be pretty widespread, so it might be a good baseline? even in armv7 cores, not sure about risc-v and powerpc though.
Jul 15, 2025
@nicbarkeragain C# makes it fairly straightforward.
Jul 15, 2025
@nicbarkeragain Intel's ISPC
Jul 15, 2025
@nicbarkeragain Llm baked into gcc who’s only job is making things simd where possible.
Jul 15, 2025
@nicbarkeragain Perhaps something like ISPC / how we author shaders in GPU could help ?
Jul 15, 2025
@nicbarkeragain Relational model + columnar data.
Jul 15, 2025
@nicbarkeragain languages that have comptime also have wonderful support for SIMD: mojo and zig
Jul 15, 2025
@nicbarkeragain Its called array programming.
Jul 15, 2025
@nicbarkeragain Maybe Futhark, or array programming.
Jul 15, 2025
@nicbarkeragain https://github.com/HigherOrderCO/Bend
Jul 15, 2025
@nicbarkeragain It hasn't been stabilized yet but Rust's portable SIMD was pretty easy to work with https://doc.rust-lang.org/std/simd/index.html
Jul 15, 2025
@nicbarkeragain more arrays
Jul 15, 2025
@nicbarkeragain openmp does good enough job with compiler flags like march=avx2, i think the focus should be making compilers smarter, for the cases where parallelism is the primary concern one should use cuda
Jul 15, 2025
@nicbarkeragain I think ISPC more or less solved this and in a C-compatible way.
Jul 15, 2025
@nicbarkeragain Might be an oversimplification but I think the focus might need to be aimed more at grouping data efficiently.
Jul 15, 2025
@nicbarkeragain Possible.
Jul 15, 2025
@nicbarkeragain So, FORTRAN?
Jul 15, 2025
@nicbarkeragain the industry is trying to do that with vector-length-agnostic architectures, like SVE, RVV, and others.
Jul 15, 2025
@nicbarkeragain I agree
Jul 15, 2025
@nicbarkeragain The "average programmer" you mentioned uses OOP and other things that make their programs so slower and fatter that SIMD would make literally zero impact within that scale of inefficiency through bad programming practices.
Jul 15, 2025
@nicbarkeragain @Modular Mojo has a first class SIMD support AFAIK but I've never used it https://docs.modular.com/mojo/stdlib/builtin/simd/SIMD#implemented-traits
Jul 15, 2025
@nicbarkeragain I agree, there is something weird with SIMD accessibility today, looks like there is room for innovation
Jul 15, 2025
@nicbarkeragain I think we need a system visualizer where devs can view their implementation running through the processor/gpus/memory/network.
Jul 15, 2025
@nicbarkeragain have you spent time writing SIMD algorithms for general-purpose tasks? it does not seem possible to make this stuff accessible, because the set of available doodads and their costs influences your algorithm design
Jul 15, 2025
@nicbarkeragain have you looked at Odin’s SIMD?
Jul 15, 2025
@nicbarkeragain Whats wrong with using inline ASM?
Jul 15, 2025
@nicbarkeragain Just enable arithmetic operations on fixed sized arrays and optimise this to use SIMD where possible
Jul 15, 2025
@nicbarkeragain Have you met the average programmer?
Jul 15, 2025
@nicbarkeragain

Created with Quartz v5.0.0 © 2026

GitHub